tm.texi revision 169690
1@c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001, 2@c 2002, 2003, 2004, 2005, 2006 Free Software Foundation, Inc. 3@c This is part of the GCC manual. 4@c For copying conditions, see the file gcc.texi. 5 6@node Target Macros 7@chapter Target Description Macros and Functions 8@cindex machine description macros 9@cindex target description macros 10@cindex macros, target description 11@cindex @file{tm.h} macros 12 13In addition to the file @file{@var{machine}.md}, a machine description 14includes a C header file conventionally given the name 15@file{@var{machine}.h} and a C source file named @file{@var{machine}.c}. 16The header file defines numerous macros that convey the information 17about the target machine that does not fit into the scheme of the 18@file{.md} file. The file @file{tm.h} should be a link to 19@file{@var{machine}.h}. The header file @file{config.h} includes 20@file{tm.h} and most compiler source files include @file{config.h}. The 21source file defines a variable @code{targetm}, which is a structure 22containing pointers to functions and data relating to the target 23machine. @file{@var{machine}.c} should also contain their definitions, 24if they are not defined elsewhere in GCC, and other functions called 25through the macros defined in the @file{.h} file. 26 27@menu 28* Target Structure:: The @code{targetm} variable. 29* Driver:: Controlling how the driver runs the compilation passes. 30* Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}. 31* Per-Function Data:: Defining data structures for per-function information. 32* Storage Layout:: Defining sizes and alignments of data. 33* Type Layout:: Defining sizes and properties of basic user data types. 34* Registers:: Naming and describing the hardware registers. 35* Register Classes:: Defining the classes of hardware registers. 36* Old Constraints:: The old way to define machine-specific constraints. 37* Stack and Calling:: Defining which way the stack grows and by how much. 38* Varargs:: Defining the varargs macros. 39* Trampolines:: Code set up at run time to enter a nested function. 40* Library Calls:: Controlling how library routines are implicitly called. 41* Addressing Modes:: Defining addressing modes valid for memory operands. 42* Anchored Addresses:: Defining how @option{-fsection-anchors} should work. 43* Condition Code:: Defining how insns update the condition code. 44* Costs:: Defining relative costs of different operations. 45* Scheduling:: Adjusting the behavior of the instruction scheduler. 46* Sections:: Dividing storage into text, data, and other sections. 47* PIC:: Macros for position independent code. 48* Assembler Format:: Defining how to write insns and pseudo-ops to output. 49* Debugging Info:: Defining the format of debugging output. 50* Floating Point:: Handling floating point for cross-compilers. 51* Mode Switching:: Insertion of mode-switching instructions. 52* Target Attributes:: Defining target-specific uses of @code{__attribute__}. 53* MIPS Coprocessors:: MIPS coprocessor support and how to customize it. 54* PCH Target:: Validity checking for precompiled headers. 55* C++ ABI:: Controlling C++ ABI changes. 56* Misc:: Everything else. 57@end menu 58 59@node Target Structure 60@section The Global @code{targetm} Variable 61@cindex target hooks 62@cindex target functions 63 64@deftypevar {struct gcc_target} targetm 65The target @file{.c} file must define the global @code{targetm} variable 66which contains pointers to functions and data relating to the target 67machine. The variable is declared in @file{target.h}; 68@file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is 69used to initialize the variable, and macros for the default initializers 70for elements of the structure. The @file{.c} file should override those 71macros for which the default definition is inappropriate. For example: 72@smallexample 73#include "target.h" 74#include "target-def.h" 75 76/* @r{Initialize the GCC target structure.} */ 77 78#undef TARGET_COMP_TYPE_ATTRIBUTES 79#define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes 80 81struct gcc_target targetm = TARGET_INITIALIZER; 82@end smallexample 83@end deftypevar 84 85Where a macro should be defined in the @file{.c} file in this manner to 86form part of the @code{targetm} structure, it is documented below as a 87``Target Hook'' with a prototype. Many macros will change in future 88from being defined in the @file{.h} file to being part of the 89@code{targetm} structure. 90 91@node Driver 92@section Controlling the Compilation Driver, @file{gcc} 93@cindex driver 94@cindex controlling the compilation driver 95 96@c prevent bad page break with this line 97You can control the compilation driver. 98 99@defmac SWITCH_TAKES_ARG (@var{char}) 100A C expression which determines whether the option @option{-@var{char}} 101takes arguments. The value should be the number of arguments that 102option takes--zero, for many options. 103 104By default, this macro is defined as 105@code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options 106properly. You need not define @code{SWITCH_TAKES_ARG} unless you 107wish to add additional options which take arguments. Any redefinition 108should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for 109additional options. 110@end defmac 111 112@defmac WORD_SWITCH_TAKES_ARG (@var{name}) 113A C expression which determines whether the option @option{-@var{name}} 114takes arguments. The value should be the number of arguments that 115option takes--zero, for many options. This macro rather than 116@code{SWITCH_TAKES_ARG} is used for multi-character option names. 117 118By default, this macro is defined as 119@code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options 120properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you 121wish to add additional options which take arguments. Any redefinition 122should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for 123additional options. 124@end defmac 125 126@defmac SWITCH_CURTAILS_COMPILATION (@var{char}) 127A C expression which determines whether the option @option{-@var{char}} 128stops compilation before the generation of an executable. The value is 129boolean, nonzero if the option does stop an executable from being 130generated, zero otherwise. 131 132By default, this macro is defined as 133@code{DEFAULT_SWITCH_CURTAILS_COMPILATION}, which handles the standard 134options properly. You need not define 135@code{SWITCH_CURTAILS_COMPILATION} unless you wish to add additional 136options which affect the generation of an executable. Any redefinition 137should call @code{DEFAULT_SWITCH_CURTAILS_COMPILATION} and then check 138for additional options. 139@end defmac 140 141@defmac SWITCHES_NEED_SPACES 142A string-valued C expression which enumerates the options for which 143the linker needs a space between the option and its argument. 144 145If this macro is not defined, the default value is @code{""}. 146@end defmac 147 148@defmac TARGET_OPTION_TRANSLATE_TABLE 149If defined, a list of pairs of strings, the first of which is a 150potential command line target to the @file{gcc} driver program, and the 151second of which is a space-separated (tabs and other whitespace are not 152supported) list of options with which to replace the first option. The 153target defining this list is responsible for assuring that the results 154are valid. Replacement options may not be the @code{--opt} style, they 155must be the @code{-opt} style. It is the intention of this macro to 156provide a mechanism for substitution that affects the multilibs chosen, 157such as one option that enables many options, some of which select 158multilibs. Example nonsensical definition, where @option{-malt-abi}, 159@option{-EB}, and @option{-mspoo} cause different multilibs to be chosen: 160 161@smallexample 162#define TARGET_OPTION_TRANSLATE_TABLE \ 163@{ "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" @}, \ 164@{ "-compat", "-EB -malign=4 -mspoo" @} 165@end smallexample 166@end defmac 167 168@defmac DRIVER_SELF_SPECS 169A list of specs for the driver itself. It should be a suitable 170initializer for an array of strings, with no surrounding braces. 171 172The driver applies these specs to its own command line between loading 173default @file{specs} files (but not command-line specified ones) and 174choosing the multilib directory or running any subcommands. It 175applies them in the order given, so each spec can depend on the 176options added by earlier ones. It is also possible to remove options 177using @samp{%<@var{option}} in the usual way. 178 179This macro can be useful when a port has several interdependent target 180options. It provides a way of standardizing the command line so 181that the other specs are easier to write. 182 183Do not define this macro if it does not need to do anything. 184@end defmac 185 186@defmac OPTION_DEFAULT_SPECS 187A list of specs used to support configure-time default options (i.e.@: 188@option{--with} options) in the driver. It should be a suitable initializer 189for an array of structures, each containing two strings, without the 190outermost pair of surrounding braces. 191 192The first item in the pair is the name of the default. This must match 193the code in @file{config.gcc} for the target. The second item is a spec 194to apply if a default with this name was specified. The string 195@samp{%(VALUE)} in the spec will be replaced by the value of the default 196everywhere it occurs. 197 198The driver will apply these specs to its own command line between loading 199default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using 200the same mechanism as @code{DRIVER_SELF_SPECS}. 201 202Do not define this macro if it does not need to do anything. 203@end defmac 204 205@defmac CPP_SPEC 206A C string constant that tells the GCC driver program options to 207pass to CPP@. It can also specify how to translate options you 208give to GCC into options for GCC to pass to the CPP@. 209 210Do not define this macro if it does not need to do anything. 211@end defmac 212 213@defmac CPLUSPLUS_CPP_SPEC 214This macro is just like @code{CPP_SPEC}, but is used for C++, rather 215than C@. If you do not define this macro, then the value of 216@code{CPP_SPEC} (if any) will be used instead. 217@end defmac 218 219@defmac CC1_SPEC 220A C string constant that tells the GCC driver program options to 221pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language 222front ends. 223It can also specify how to translate options you give to GCC into options 224for GCC to pass to front ends. 225 226Do not define this macro if it does not need to do anything. 227@end defmac 228 229@defmac CC1PLUS_SPEC 230A C string constant that tells the GCC driver program options to 231pass to @code{cc1plus}. It can also specify how to translate options you 232give to GCC into options for GCC to pass to the @code{cc1plus}. 233 234Do not define this macro if it does not need to do anything. 235Note that everything defined in CC1_SPEC is already passed to 236@code{cc1plus} so there is no need to duplicate the contents of 237CC1_SPEC in CC1PLUS_SPEC@. 238@end defmac 239 240@defmac ASM_SPEC 241A C string constant that tells the GCC driver program options to 242pass to the assembler. It can also specify how to translate options 243you give to GCC into options for GCC to pass to the assembler. 244See the file @file{sun3.h} for an example of this. 245 246Do not define this macro if it does not need to do anything. 247@end defmac 248 249@defmac ASM_FINAL_SPEC 250A C string constant that tells the GCC driver program how to 251run any programs which cleanup after the normal assembler. 252Normally, this is not needed. See the file @file{mips.h} for 253an example of this. 254 255Do not define this macro if it does not need to do anything. 256@end defmac 257 258@defmac AS_NEEDS_DASH_FOR_PIPED_INPUT 259Define this macro, with no value, if the driver should give the assembler 260an argument consisting of a single dash, @option{-}, to instruct it to 261read from its standard input (which will be a pipe connected to the 262output of the compiler proper). This argument is given after any 263@option{-o} option specifying the name of the output file. 264 265If you do not define this macro, the assembler is assumed to read its 266standard input if given no non-option arguments. If your assembler 267cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct; 268see @file{mips.h} for instance. 269@end defmac 270 271@defmac LINK_SPEC 272A C string constant that tells the GCC driver program options to 273pass to the linker. It can also specify how to translate options you 274give to GCC into options for GCC to pass to the linker. 275 276Do not define this macro if it does not need to do anything. 277@end defmac 278 279@defmac LIB_SPEC 280Another C string constant used much like @code{LINK_SPEC}. The difference 281between the two is that @code{LIB_SPEC} is used at the end of the 282command given to the linker. 283 284If this macro is not defined, a default is provided that 285loads the standard C library from the usual place. See @file{gcc.c}. 286@end defmac 287 288@defmac LIBGCC_SPEC 289Another C string constant that tells the GCC driver program 290how and when to place a reference to @file{libgcc.a} into the 291linker command line. This constant is placed both before and after 292the value of @code{LIB_SPEC}. 293 294If this macro is not defined, the GCC driver provides a default that 295passes the string @option{-lgcc} to the linker. 296@end defmac 297 298@defmac REAL_LIBGCC_SPEC 299By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the 300@code{LIBGCC_SPEC} is not directly used by the driver program but is 301instead modified to refer to different versions of @file{libgcc.a} 302depending on the values of the command line flags @option{-static}, 303@option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On 304targets where these modifications are inappropriate, define 305@code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the 306driver how to place a reference to @file{libgcc} on the link command 307line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified. 308@end defmac 309 310@defmac USE_LD_AS_NEEDED 311A macro that controls the modifications to @code{LIBGCC_SPEC} 312mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be 313generated that uses --as-needed and the shared libgcc in place of the 314static exception handler library, when linking without any of 315@code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}. 316@end defmac 317 318@defmac LINK_EH_SPEC 319If defined, this C string constant is added to @code{LINK_SPEC}. 320When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects 321the modifications to @code{LIBGCC_SPEC} mentioned in 322@code{REAL_LIBGCC_SPEC}. 323@end defmac 324 325@defmac STARTFILE_SPEC 326Another C string constant used much like @code{LINK_SPEC}. The 327difference between the two is that @code{STARTFILE_SPEC} is used at 328the very beginning of the command given to the linker. 329 330If this macro is not defined, a default is provided that loads the 331standard C startup file from the usual place. See @file{gcc.c}. 332@end defmac 333 334@defmac ENDFILE_SPEC 335Another C string constant used much like @code{LINK_SPEC}. The 336difference between the two is that @code{ENDFILE_SPEC} is used at 337the very end of the command given to the linker. 338 339Do not define this macro if it does not need to do anything. 340@end defmac 341 342@defmac THREAD_MODEL_SPEC 343GCC @code{-v} will print the thread model GCC was configured to use. 344However, this doesn't work on platforms that are multilibbed on thread 345models, such as AIX 4.3. On such platforms, define 346@code{THREAD_MODEL_SPEC} such that it evaluates to a string without 347blanks that names one of the recognized thread models. @code{%*}, the 348default value of this macro, will expand to the value of 349@code{thread_file} set in @file{config.gcc}. 350@end defmac 351 352@defmac SYSROOT_SUFFIX_SPEC 353Define this macro to add a suffix to the target sysroot when GCC is 354configured with a sysroot. This will cause GCC to search for usr/lib, 355et al, within sysroot+suffix. 356@end defmac 357 358@defmac SYSROOT_HEADERS_SUFFIX_SPEC 359Define this macro to add a headers_suffix to the target sysroot when 360GCC is configured with a sysroot. This will cause GCC to pass the 361updated sysroot+headers_suffix to CPP, causing it to search for 362usr/include, et al, within sysroot+headers_suffix. 363@end defmac 364 365@defmac EXTRA_SPECS 366Define this macro to provide additional specifications to put in the 367@file{specs} file that can be used in various specifications like 368@code{CC1_SPEC}. 369 370The definition should be an initializer for an array of structures, 371containing a string constant, that defines the specification name, and a 372string constant that provides the specification. 373 374Do not define this macro if it does not need to do anything. 375 376@code{EXTRA_SPECS} is useful when an architecture contains several 377related targets, which have various @code{@dots{}_SPECS} which are similar 378to each other, and the maintainer would like one central place to keep 379these definitions. 380 381For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to 382define either @code{_CALL_SYSV} when the System V calling sequence is 383used or @code{_CALL_AIX} when the older AIX-based calling sequence is 384used. 385 386The @file{config/rs6000/rs6000.h} target file defines: 387 388@smallexample 389#define EXTRA_SPECS \ 390 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @}, 391 392#define CPP_SYS_DEFAULT "" 393@end smallexample 394 395The @file{config/rs6000/sysv.h} target file defines: 396@smallexample 397#undef CPP_SPEC 398#define CPP_SPEC \ 399"%@{posix: -D_POSIX_SOURCE @} \ 400%@{mcall-sysv: -D_CALL_SYSV @} \ 401%@{!mcall-sysv: %(cpp_sysv_default) @} \ 402%@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}" 403 404#undef CPP_SYSV_DEFAULT 405#define CPP_SYSV_DEFAULT "-D_CALL_SYSV" 406@end smallexample 407 408while the @file{config/rs6000/eabiaix.h} target file defines 409@code{CPP_SYSV_DEFAULT} as: 410 411@smallexample 412#undef CPP_SYSV_DEFAULT 413#define CPP_SYSV_DEFAULT "-D_CALL_AIX" 414@end smallexample 415@end defmac 416 417@defmac LINK_LIBGCC_SPECIAL_1 418Define this macro if the driver program should find the library 419@file{libgcc.a}. If you do not define this macro, the driver program will pass 420the argument @option{-lgcc} to tell the linker to do the search. 421@end defmac 422 423@defmac LINK_GCC_C_SEQUENCE_SPEC 424The sequence in which libgcc and libc are specified to the linker. 425By default this is @code{%G %L %G}. 426@end defmac 427 428@defmac LINK_COMMAND_SPEC 429A C string constant giving the complete command line need to execute the 430linker. When you do this, you will need to update your port each time a 431change is made to the link command line within @file{gcc.c}. Therefore, 432define this macro only if you need to completely redefine the command 433line for invoking the linker and there is no other way to accomplish 434the effect you need. Overriding this macro may be avoidable by overriding 435@code{LINK_GCC_C_SEQUENCE_SPEC} instead. 436@end defmac 437 438@defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES 439A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search 440directories from linking commands. Do not give it a nonzero value if 441removing duplicate search directories changes the linker's semantics. 442@end defmac 443 444@defmac MULTILIB_DEFAULTS 445Define this macro as a C expression for the initializer of an array of 446string to tell the driver program which options are defaults for this 447target and thus do not need to be handled specially when using 448@code{MULTILIB_OPTIONS}. 449 450Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in 451the target makefile fragment or if none of the options listed in 452@code{MULTILIB_OPTIONS} are set by default. 453@xref{Target Fragment}. 454@end defmac 455 456@defmac RELATIVE_PREFIX_NOT_LINKDIR 457Define this macro to tell @command{gcc} that it should only translate 458a @option{-B} prefix into a @option{-L} linker option if the prefix 459indicates an absolute file name. 460@end defmac 461 462@defmac MD_EXEC_PREFIX 463If defined, this macro is an additional prefix to try after 464@code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched 465when the @option{-b} option is used, or the compiler is built as a cross 466compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it 467to the list of directories used to find the assembler in @file{configure.in}. 468@end defmac 469 470@defmac STANDARD_STARTFILE_PREFIX 471Define this macro as a C string constant if you wish to override the 472standard choice of @code{libdir} as the default prefix to 473try when searching for startup files such as @file{crt0.o}. 474@code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler 475is built as a cross compiler. 476@end defmac 477 478@defmac STANDARD_STARTFILE_PREFIX_1 479Define this macro as a C string constant if you wish to override the 480standard choice of @code{/lib} as a prefix to try after the default prefix 481when searching for startup files such as @file{crt0.o}. 482@code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler 483is built as a cross compiler. 484@end defmac 485 486@defmac STANDARD_STARTFILE_PREFIX_2 487Define this macro as a C string constant if you wish to override the 488standard choice of @code{/lib} as yet another prefix to try after the 489default prefix when searching for startup files such as @file{crt0.o}. 490@code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler 491is built as a cross compiler. 492@end defmac 493 494@defmac MD_STARTFILE_PREFIX 495If defined, this macro supplies an additional prefix to try after the 496standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the 497@option{-b} option is used, or when the compiler is built as a cross 498compiler. 499@end defmac 500 501@defmac MD_STARTFILE_PREFIX_1 502If defined, this macro supplies yet another prefix to try after the 503standard prefixes. It is not searched when the @option{-b} option is 504used, or when the compiler is built as a cross compiler. 505@end defmac 506 507@defmac INIT_ENVIRONMENT 508Define this macro as a C string constant if you wish to set environment 509variables for programs called by the driver, such as the assembler and 510loader. The driver passes the value of this macro to @code{putenv} to 511initialize the necessary environment variables. 512@end defmac 513 514@defmac LOCAL_INCLUDE_DIR 515Define this macro as a C string constant if you wish to override the 516standard choice of @file{/usr/local/include} as the default prefix to 517try when searching for local header files. @code{LOCAL_INCLUDE_DIR} 518comes before @code{SYSTEM_INCLUDE_DIR} in the search order. 519 520Cross compilers do not search either @file{/usr/local/include} or its 521replacement. 522@end defmac 523 524@defmac MODIFY_TARGET_NAME 525Define this macro if you wish to define command-line switches that 526modify the default target name. 527 528For each switch, you can include a string to be appended to the first 529part of the configuration name or a string to be deleted from the 530configuration name, if present. The definition should be an initializer 531for an array of structures. Each array element should have three 532elements: the switch name (a string constant, including the initial 533dash), one of the enumeration codes @code{ADD} or @code{DELETE} to 534indicate whether the string should be inserted or deleted, and the string 535to be inserted or deleted (a string constant). 536 537For example, on a machine where @samp{64} at the end of the 538configuration name denotes a 64-bit target and you want the @option{-32} 539and @option{-64} switches to select between 32- and 64-bit targets, you would 540code 541 542@smallexample 543#define MODIFY_TARGET_NAME \ 544 @{ @{ "-32", DELETE, "64"@}, \ 545 @{"-64", ADD, "64"@}@} 546@end smallexample 547@end defmac 548 549@defmac SYSTEM_INCLUDE_DIR 550Define this macro as a C string constant if you wish to specify a 551system-specific directory to search for header files before the standard 552directory. @code{SYSTEM_INCLUDE_DIR} comes before 553@code{STANDARD_INCLUDE_DIR} in the search order. 554 555Cross compilers do not use this macro and do not search the directory 556specified. 557@end defmac 558 559@defmac STANDARD_INCLUDE_DIR 560Define this macro as a C string constant if you wish to override the 561standard choice of @file{/usr/include} as the default prefix to 562try when searching for header files. 563 564Cross compilers ignore this macro and do not search either 565@file{/usr/include} or its replacement. 566@end defmac 567 568@defmac STANDARD_INCLUDE_COMPONENT 569The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}. 570See @code{INCLUDE_DEFAULTS}, below, for the description of components. 571If you do not define this macro, no component is used. 572@end defmac 573 574@defmac INCLUDE_DEFAULTS 575Define this macro if you wish to override the entire default search path 576for include files. For a native compiler, the default search path 577usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR}, 578@code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and 579@code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR} 580and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile}, 581and specify private search areas for GCC@. The directory 582@code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs. 583 584The definition should be an initializer for an array of structures. 585Each array element should have four elements: the directory name (a 586string constant), the component name (also a string constant), a flag 587for C++-only directories, 588and a flag showing that the includes in the directory don't need to be 589wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of 590the array with a null element. 591 592The component name denotes what GNU package the include file is part of, 593if any, in all uppercase letters. For example, it might be @samp{GCC} 594or @samp{BINUTILS}. If the package is part of a vendor-supplied 595operating system, code the component name as @samp{0}. 596 597For example, here is the definition used for VAX/VMS: 598 599@smallexample 600#define INCLUDE_DEFAULTS \ 601@{ \ 602 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \ 603 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \ 604 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \ 605 @{ ".", 0, 0, 0@}, \ 606 @{ 0, 0, 0, 0@} \ 607@} 608@end smallexample 609@end defmac 610 611Here is the order of prefixes tried for exec files: 612 613@enumerate 614@item 615Any prefixes specified by the user with @option{-B}. 616 617@item 618The environment variable @code{GCC_EXEC_PREFIX}, if any. 619 620@item 621The directories specified by the environment variable @code{COMPILER_PATH}. 622 623@item 624The macro @code{STANDARD_EXEC_PREFIX}. 625 626@item 627@file{/usr/lib/gcc/}. 628 629@item 630The macro @code{MD_EXEC_PREFIX}, if any. 631@end enumerate 632 633Here is the order of prefixes tried for startfiles: 634 635@enumerate 636@item 637Any prefixes specified by the user with @option{-B}. 638 639@item 640The environment variable @code{GCC_EXEC_PREFIX}, if any. 641 642@item 643The directories specified by the environment variable @code{LIBRARY_PATH} 644(or port-specific name; native only, cross compilers do not use this). 645 646@item 647The macro @code{STANDARD_EXEC_PREFIX}. 648 649@item 650@file{/usr/lib/gcc/}. 651 652@item 653The macro @code{MD_EXEC_PREFIX}, if any. 654 655@item 656The macro @code{MD_STARTFILE_PREFIX}, if any. 657 658@item 659The macro @code{STANDARD_STARTFILE_PREFIX}. 660 661@item 662@file{/lib/}. 663 664@item 665@file{/usr/lib/}. 666@end enumerate 667 668@node Run-time Target 669@section Run-time Target Specification 670@cindex run-time target specification 671@cindex predefined macros 672@cindex target specifications 673 674@c prevent bad page break with this line 675Here are run-time target specifications. 676 677@defmac TARGET_CPU_CPP_BUILTINS () 678This function-like macro expands to a block of code that defines 679built-in preprocessor macros and assertions for the target cpu, using 680the functions @code{builtin_define}, @code{builtin_define_std} and 681@code{builtin_assert}. When the front end 682calls this macro it provides a trailing semicolon, and since it has 683finished command line option processing your code can use those 684results freely. 685 686@code{builtin_assert} takes a string in the form you pass to the 687command-line option @option{-A}, such as @code{cpu=mips}, and creates 688the assertion. @code{builtin_define} takes a string in the form 689accepted by option @option{-D} and unconditionally defines the macro. 690 691@code{builtin_define_std} takes a string representing the name of an 692object-like macro. If it doesn't lie in the user's namespace, 693@code{builtin_define_std} defines it unconditionally. Otherwise, it 694defines a version with two leading underscores, and another version 695with two leading and trailing underscores, and defines the original 696only if an ISO standard was not requested on the command line. For 697example, passing @code{unix} defines @code{__unix}, @code{__unix__} 698and possibly @code{unix}; passing @code{_mips} defines @code{__mips}, 699@code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64} 700defines only @code{_ABI64}. 701 702You can also test for the C dialect being compiled. The variable 703@code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus} 704or @code{clk_objective_c}. Note that if we are preprocessing 705assembler, this variable will be @code{clk_c} but the function-like 706macro @code{preprocessing_asm_p()} will return true, so you might want 707to check for that first. If you need to check for strict ANSI, the 708variable @code{flag_iso} can be used. The function-like macro 709@code{preprocessing_trad_p()} can be used to check for traditional 710preprocessing. 711@end defmac 712 713@defmac TARGET_OS_CPP_BUILTINS () 714Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional 715and is used for the target operating system instead. 716@end defmac 717 718@defmac TARGET_OBJFMT_CPP_BUILTINS () 719Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional 720and is used for the target object format. @file{elfos.h} uses this 721macro to define @code{__ELF__}, so you probably do not need to define 722it yourself. 723@end defmac 724 725@deftypevar {extern int} target_flags 726This variable is declared in @file{options.h}, which is included before 727any target-specific headers. 728@end deftypevar 729 730@deftypevar {Target Hook} int TARGET_DEFAULT_TARGET_FLAGS 731This variable specifies the initial value of @code{target_flags}. 732Its default setting is 0. 733@end deftypevar 734 735@cindex optional hardware or system features 736@cindex features, optional, in system conventions 737 738@deftypefn {Target Hook} bool TARGET_HANDLE_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value}) 739This hook is called whenever the user specifies one of the 740target-specific options described by the @file{.opt} definition files 741(@pxref{Options}). It has the opportunity to do some option-specific 742processing and should return true if the option is valid. The default 743definition does nothing but return true. 744 745@var{code} specifies the @code{OPT_@var{name}} enumeration value 746associated with the selected option; @var{name} is just a rendering of 747the option name in which non-alphanumeric characters are replaced by 748underscores. @var{arg} specifies the string argument and is null if 749no argument was given. If the option is flagged as a @code{UInteger} 750(@pxref{Option properties}), @var{value} is the numeric value of the 751argument. Otherwise @var{value} is 1 if the positive form of the 752option was used and 0 if the ``no-'' form was. 753@end deftypefn 754 755@defmac TARGET_VERSION 756This macro is a C statement to print on @code{stderr} a string 757describing the particular machine description choice. Every machine 758description should define @code{TARGET_VERSION}. For example: 759 760@smallexample 761#ifdef MOTOROLA 762#define TARGET_VERSION \ 763 fprintf (stderr, " (68k, Motorola syntax)"); 764#else 765#define TARGET_VERSION \ 766 fprintf (stderr, " (68k, MIT syntax)"); 767#endif 768@end smallexample 769@end defmac 770 771@defmac OVERRIDE_OPTIONS 772Sometimes certain combinations of command options do not make sense on 773a particular target machine. You can define a macro 774@code{OVERRIDE_OPTIONS} to take account of this. This macro, if 775defined, is executed once just after all the command options have been 776parsed. 777 778Don't use this macro to turn on various extra optimizations for 779@option{-O}. That is what @code{OPTIMIZATION_OPTIONS} is for. 780@end defmac 781 782@defmac C_COMMON_OVERRIDE_OPTIONS 783This is similar to @code{OVERRIDE_OPTIONS} but is only used in the C 784language frontends (C, Objective-C, C++, Objective-C++) and so can be 785used to alter option flag variables which only exist in those 786frontends. 787@end defmac 788 789@defmac OPTIMIZATION_OPTIONS (@var{level}, @var{size}) 790Some machines may desire to change what optimizations are performed for 791various optimization levels. This macro, if defined, is executed once 792just after the optimization level is determined and before the remainder 793of the command options have been parsed. Values set in this macro are 794used as the default values for the other command line options. 795 796@var{level} is the optimization level specified; 2 if @option{-O2} is 797specified, 1 if @option{-O} is specified, and 0 if neither is specified. 798 799@var{size} is nonzero if @option{-Os} is specified and zero otherwise. 800 801You should not use this macro to change options that are not 802machine-specific. These should uniformly selected by the same 803optimization level on all supported machines. Use this macro to enable 804machine-specific optimizations. 805 806@strong{Do not examine @code{write_symbols} in 807this macro!} The debugging options are not supposed to alter the 808generated code. 809@end defmac 810 811@defmac CAN_DEBUG_WITHOUT_FP 812Define this macro if debugging can be performed even without a frame 813pointer. If this macro is defined, GCC will turn on the 814@option{-fomit-frame-pointer} option whenever @option{-O} is specified. 815@end defmac 816 817@node Per-Function Data 818@section Defining data structures for per-function information. 819@cindex per-function data 820@cindex data structures 821 822If the target needs to store information on a per-function basis, GCC 823provides a macro and a couple of variables to allow this. Note, just 824using statics to store the information is a bad idea, since GCC supports 825nested functions, so you can be halfway through encoding one function 826when another one comes along. 827 828GCC defines a data structure called @code{struct function} which 829contains all of the data specific to an individual function. This 830structure contains a field called @code{machine} whose type is 831@code{struct machine_function *}, which can be used by targets to point 832to their own specific data. 833 834If a target needs per-function specific data it should define the type 835@code{struct machine_function} and also the macro @code{INIT_EXPANDERS}. 836This macro should be used to initialize the function pointer 837@code{init_machine_status}. This pointer is explained below. 838 839One typical use of per-function, target specific data is to create an 840RTX to hold the register containing the function's return address. This 841RTX can then be used to implement the @code{__builtin_return_address} 842function, for level 0. 843 844Note---earlier implementations of GCC used a single data area to hold 845all of the per-function information. Thus when processing of a nested 846function began the old per-function data had to be pushed onto a 847stack, and when the processing was finished, it had to be popped off the 848stack. GCC used to provide function pointers called 849@code{save_machine_status} and @code{restore_machine_status} to handle 850the saving and restoring of the target specific information. Since the 851single data area approach is no longer used, these pointers are no 852longer supported. 853 854@defmac INIT_EXPANDERS 855Macro called to initialize any target specific information. This macro 856is called once per function, before generation of any RTL has begun. 857The intention of this macro is to allow the initialization of the 858function pointer @code{init_machine_status}. 859@end defmac 860 861@deftypevar {void (*)(struct function *)} init_machine_status 862If this function pointer is non-@code{NULL} it will be called once per 863function, before function compilation starts, in order to allow the 864target to perform any target specific initialization of the 865@code{struct function} structure. It is intended that this would be 866used to initialize the @code{machine} of that structure. 867 868@code{struct machine_function} structures are expected to be freed by GC@. 869Generally, any memory that they reference must be allocated by using 870@code{ggc_alloc}, including the structure itself. 871@end deftypevar 872 873@node Storage Layout 874@section Storage Layout 875@cindex storage layout 876 877Note that the definitions of the macros in this table which are sizes or 878alignments measured in bits do not need to be constant. They can be C 879expressions that refer to static variables, such as the @code{target_flags}. 880@xref{Run-time Target}. 881 882@defmac BITS_BIG_ENDIAN 883Define this macro to have the value 1 if the most significant bit in a 884byte has the lowest number; otherwise define it to have the value zero. 885This means that bit-field instructions count from the most significant 886bit. If the machine has no bit-field instructions, then this must still 887be defined, but it doesn't matter which value it is defined to. This 888macro need not be a constant. 889 890This macro does not affect the way structure fields are packed into 891bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}. 892@end defmac 893 894@defmac BYTES_BIG_ENDIAN 895Define this macro to have the value 1 if the most significant byte in a 896word has the lowest number. This macro need not be a constant. 897@end defmac 898 899@defmac WORDS_BIG_ENDIAN 900Define this macro to have the value 1 if, in a multiword object, the 901most significant word has the lowest number. This applies to both 902memory locations and registers; GCC fundamentally assumes that the 903order of words in memory is the same as the order in registers. This 904macro need not be a constant. 905@end defmac 906 907@defmac LIBGCC2_WORDS_BIG_ENDIAN 908Define this macro if @code{WORDS_BIG_ENDIAN} is not constant. This must be a 909constant value with the same meaning as @code{WORDS_BIG_ENDIAN}, which will be 910used only when compiling @file{libgcc2.c}. Typically the value will be set 911based on preprocessor defines. 912@end defmac 913 914@defmac FLOAT_WORDS_BIG_ENDIAN 915Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or 916@code{TFmode} floating point numbers are stored in memory with the word 917containing the sign bit at the lowest address; otherwise define it to 918have the value 0. This macro need not be a constant. 919 920You need not define this macro if the ordering is the same as for 921multi-word integers. 922@end defmac 923 924@defmac BITS_PER_UNIT 925Define this macro to be the number of bits in an addressable storage 926unit (byte). If you do not define this macro the default is 8. 927@end defmac 928 929@defmac BITS_PER_WORD 930Number of bits in a word. If you do not define this macro, the default 931is @code{BITS_PER_UNIT * UNITS_PER_WORD}. 932@end defmac 933 934@defmac MAX_BITS_PER_WORD 935Maximum number of bits in a word. If this is undefined, the default is 936@code{BITS_PER_WORD}. Otherwise, it is the constant value that is the 937largest value that @code{BITS_PER_WORD} can have at run-time. 938@end defmac 939 940@defmac UNITS_PER_WORD 941Number of storage units in a word; normally the size of a general-purpose 942register, a power of two from 1 or 8. 943@end defmac 944 945@defmac MIN_UNITS_PER_WORD 946Minimum number of units in a word. If this is undefined, the default is 947@code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the 948smallest value that @code{UNITS_PER_WORD} can have at run-time. 949@end defmac 950 951@defmac UNITS_PER_SIMD_WORD 952Number of units in the vectors that the vectorizer can produce. 953The default is equal to @code{UNITS_PER_WORD}, because the vectorizer 954can do some transformations even in absence of specialized @acronym{SIMD} 955hardware. 956@end defmac 957 958@defmac POINTER_SIZE 959Width of a pointer, in bits. You must specify a value no wider than the 960width of @code{Pmode}. If it is not equal to the width of @code{Pmode}, 961you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify 962a value the default is @code{BITS_PER_WORD}. 963@end defmac 964 965@defmac POINTERS_EXTEND_UNSIGNED 966A C expression whose value is greater than zero if pointers that need to be 967extended from being @code{POINTER_SIZE} bits wide to @code{Pmode} are to 968be zero-extended and zero if they are to be sign-extended. If the value 969is less then zero then there must be an "ptr_extend" instruction that 970extends a pointer from @code{POINTER_SIZE} to @code{Pmode}. 971 972You need not define this macro if the @code{POINTER_SIZE} is equal 973to the width of @code{Pmode}. 974@end defmac 975 976@defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type}) 977A macro to update @var{m} and @var{unsignedp} when an object whose type 978is @var{type} and which has the specified mode and signedness is to be 979stored in a register. This macro is only called when @var{type} is a 980scalar type. 981 982On most RISC machines, which only have operations that operate on a full 983register, define this macro to set @var{m} to @code{word_mode} if 984@var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most 985cases, only integer modes should be widened because wider-precision 986floating-point operations are usually more expensive than their narrower 987counterparts. 988 989For most machines, the macro definition does not change @var{unsignedp}. 990However, some machines, have instructions that preferentially handle 991either signed or unsigned quantities of certain modes. For example, on 992the DEC Alpha, 32-bit loads from memory and 32-bit add instructions 993sign-extend the result to 64 bits. On such machines, set 994@var{unsignedp} according to which kind of extension is more efficient. 995 996Do not define this macro if it would never modify @var{m}. 997@end defmac 998 999@defmac PROMOTE_FUNCTION_MODE 1000Like @code{PROMOTE_MODE}, but is applied to outgoing function arguments or 1001function return values, as specified by @code{TARGET_PROMOTE_FUNCTION_ARGS} 1002and @code{TARGET_PROMOTE_FUNCTION_RETURN}, respectively. 1003 1004The default is @code{PROMOTE_MODE}. 1005@end defmac 1006 1007@deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_ARGS (tree @var{fntype}) 1008This target hook should return @code{true} if the promotion described by 1009@code{PROMOTE_FUNCTION_MODE} should be done for outgoing function 1010arguments. 1011@end deftypefn 1012 1013@deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_RETURN (tree @var{fntype}) 1014This target hook should return @code{true} if the promotion described by 1015@code{PROMOTE_FUNCTION_MODE} should be done for the return value of 1016functions. 1017 1018If this target hook returns @code{true}, @code{TARGET_FUNCTION_VALUE} 1019must perform the same promotions done by @code{PROMOTE_FUNCTION_MODE}. 1020@end deftypefn 1021 1022@defmac PARM_BOUNDARY 1023Normal alignment required for function parameters on the stack, in 1024bits. All stack parameters receive at least this much alignment 1025regardless of data type. On most machines, this is the same as the 1026size of an integer. 1027@end defmac 1028 1029@defmac STACK_BOUNDARY 1030Define this macro to the minimum alignment enforced by hardware for the 1031stack pointer on this machine. The definition is a C expression for the 1032desired alignment (measured in bits). This value is used as a default 1033if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines, 1034this should be the same as @code{PARM_BOUNDARY}. 1035@end defmac 1036 1037@defmac PREFERRED_STACK_BOUNDARY 1038Define this macro if you wish to preserve a certain alignment for the 1039stack pointer, greater than what the hardware enforces. The definition 1040is a C expression for the desired alignment (measured in bits). This 1041macro must evaluate to a value equal to or larger than 1042@code{STACK_BOUNDARY}. 1043@end defmac 1044 1045@defmac FUNCTION_BOUNDARY 1046Alignment required for a function entry point, in bits. 1047@end defmac 1048 1049@defmac BIGGEST_ALIGNMENT 1050Biggest alignment that any data type can require on this machine, in bits. 1051@end defmac 1052 1053@defmac MINIMUM_ATOMIC_ALIGNMENT 1054If defined, the smallest alignment, in bits, that can be given to an 1055object that can be referenced in one operation, without disturbing any 1056nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger 1057on machines that don't have byte or half-word store operations. 1058@end defmac 1059 1060@defmac BIGGEST_FIELD_ALIGNMENT 1061Biggest alignment that any structure or union field can require on this 1062machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for 1063structure and union fields only, unless the field alignment has been set 1064by the @code{__attribute__ ((aligned (@var{n})))} construct. 1065@end defmac 1066 1067@defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed}) 1068An expression for the alignment of a structure field @var{field} if the 1069alignment computed in the usual way (including applying of 1070@code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the 1071alignment) is @var{computed}. It overrides alignment only if the 1072field alignment has not been set by the 1073@code{__attribute__ ((aligned (@var{n})))} construct. 1074@end defmac 1075 1076@defmac MAX_OFILE_ALIGNMENT 1077Biggest alignment supported by the object file format of this machine. 1078Use this macro to limit the alignment which can be specified using the 1079@code{__attribute__ ((aligned (@var{n})))} construct. If not defined, 1080the default value is @code{BIGGEST_ALIGNMENT}. 1081@end defmac 1082 1083@defmac DATA_ALIGNMENT (@var{type}, @var{basic-align}) 1084If defined, a C expression to compute the alignment for a variable in 1085the static store. @var{type} is the data type, and @var{basic-align} is 1086the alignment that the object would ordinarily have. The value of this 1087macro is used instead of that alignment to align the object. 1088 1089If this macro is not defined, then @var{basic-align} is used. 1090 1091@findex strcpy 1092One use of this macro is to increase alignment of medium-size data to 1093make it all fit in fewer cache lines. Another is to cause character 1094arrays to be word-aligned so that @code{strcpy} calls that copy 1095constants to character arrays can be done inline. 1096@end defmac 1097 1098@defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align}) 1099If defined, a C expression to compute the alignment given to a constant 1100that is being placed in memory. @var{constant} is the constant and 1101@var{basic-align} is the alignment that the object would ordinarily 1102have. The value of this macro is used instead of that alignment to 1103align the object. 1104 1105If this macro is not defined, then @var{basic-align} is used. 1106 1107The typical use of this macro is to increase alignment for string 1108constants to be word aligned so that @code{strcpy} calls that copy 1109constants can be done inline. 1110@end defmac 1111 1112@defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align}) 1113If defined, a C expression to compute the alignment for a variable in 1114the local store. @var{type} is the data type, and @var{basic-align} is 1115the alignment that the object would ordinarily have. The value of this 1116macro is used instead of that alignment to align the object. 1117 1118If this macro is not defined, then @var{basic-align} is used. 1119 1120One use of this macro is to increase alignment of medium-size data to 1121make it all fit in fewer cache lines. 1122@end defmac 1123 1124@defmac EMPTY_FIELD_BOUNDARY 1125Alignment in bits to be given to a structure bit-field that follows an 1126empty field such as @code{int : 0;}. 1127 1128If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro. 1129@end defmac 1130 1131@defmac STRUCTURE_SIZE_BOUNDARY 1132Number of bits which any structure or union's size must be a multiple of. 1133Each structure or union's size is rounded up to a multiple of this. 1134 1135If you do not define this macro, the default is the same as 1136@code{BITS_PER_UNIT}. 1137@end defmac 1138 1139@defmac STRICT_ALIGNMENT 1140Define this macro to be the value 1 if instructions will fail to work 1141if given data not on the nominal alignment. If instructions will merely 1142go slower in that case, define this macro as 0. 1143@end defmac 1144 1145@defmac PCC_BITFIELD_TYPE_MATTERS 1146Define this if you wish to imitate the way many other C compilers handle 1147alignment of bit-fields and the structures that contain them. 1148 1149The behavior is that the type written for a named bit-field (@code{int}, 1150@code{short}, or other integer type) imposes an alignment for the entire 1151structure, as if the structure really did contain an ordinary field of 1152that type. In addition, the bit-field is placed within the structure so 1153that it would fit within such a field, not crossing a boundary for it. 1154 1155Thus, on most machines, a named bit-field whose type is written as 1156@code{int} would not cross a four-byte boundary, and would force 1157four-byte alignment for the whole structure. (The alignment used may 1158not be four bytes; it is controlled by the other alignment parameters.) 1159 1160An unnamed bit-field will not affect the alignment of the containing 1161structure. 1162 1163If the macro is defined, its definition should be a C expression; 1164a nonzero value for the expression enables this behavior. 1165 1166Note that if this macro is not defined, or its value is zero, some 1167bit-fields may cross more than one alignment boundary. The compiler can 1168support such references if there are @samp{insv}, @samp{extv}, and 1169@samp{extzv} insns that can directly reference memory. 1170 1171The other known way of making bit-fields work is to define 1172@code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}. 1173Then every structure can be accessed with fullwords. 1174 1175Unless the machine has bit-field instructions or you define 1176@code{STRUCTURE_SIZE_BOUNDARY} that way, you must define 1177@code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value. 1178 1179If your aim is to make GCC use the same conventions for laying out 1180bit-fields as are used by another compiler, here is how to investigate 1181what the other compiler does. Compile and run this program: 1182 1183@smallexample 1184struct foo1 1185@{ 1186 char x; 1187 char :0; 1188 char y; 1189@}; 1190 1191struct foo2 1192@{ 1193 char x; 1194 int :0; 1195 char y; 1196@}; 1197 1198main () 1199@{ 1200 printf ("Size of foo1 is %d\n", 1201 sizeof (struct foo1)); 1202 printf ("Size of foo2 is %d\n", 1203 sizeof (struct foo2)); 1204 exit (0); 1205@} 1206@end smallexample 1207 1208If this prints 2 and 5, then the compiler's behavior is what you would 1209get from @code{PCC_BITFIELD_TYPE_MATTERS}. 1210@end defmac 1211 1212@defmac BITFIELD_NBYTES_LIMITED 1213Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited 1214to aligning a bit-field within the structure. 1215@end defmac 1216 1217@deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELDS (void) 1218When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine 1219whether unnamed bitfields affect the alignment of the containing 1220structure. The hook should return true if the structure should inherit 1221the alignment requirements of an unnamed bitfield's type. 1222@end deftypefn 1223 1224@deftypefn {Target Hook} bool TARGET_NARROW_VOLATILE_BITFIELDS (void) 1225This target hook should return @code{true} if accesses to volatile bitfields 1226should use the narrowest mode possible. It should return @code{false} if 1227these accesses should use the bitfield container type. 1228 1229The default is @code{!TARGET_STRICT_ALIGN}. 1230@end deftypefn 1231 1232@defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode}) 1233Return 1 if a structure or array containing @var{field} should be accessed using 1234@code{BLKMODE}. 1235 1236If @var{field} is the only field in the structure, @var{mode} is its 1237mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the 1238case where structures of one field would require the structure's mode to 1239retain the field's mode. 1240 1241Normally, this is not needed. See the file @file{c4x.h} for an example 1242of how to use this macro to prevent a structure having a floating point 1243field from being accessed in an integer mode. 1244@end defmac 1245 1246@defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified}) 1247Define this macro as an expression for the alignment of a type (given 1248by @var{type} as a tree node) if the alignment computed in the usual 1249way is @var{computed} and the alignment explicitly specified was 1250@var{specified}. 1251 1252The default is to use @var{specified} if it is larger; otherwise, use 1253the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT} 1254@end defmac 1255 1256@defmac MAX_FIXED_MODE_SIZE 1257An integer expression for the size in bits of the largest integer 1258machine mode that should actually be used. All integer machine modes of 1259this size or smaller can be used for structures and unions with the 1260appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE 1261(DImode)} is assumed. 1262@end defmac 1263 1264@defmac STACK_SAVEAREA_MODE (@var{save_level}) 1265If defined, an expression of type @code{enum machine_mode} that 1266specifies the mode of the save area operand of a 1267@code{save_stack_@var{level}} named pattern (@pxref{Standard Names}). 1268@var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or 1269@code{SAVE_NONLOCAL} and selects which of the three named patterns is 1270having its mode specified. 1271 1272You need not define this macro if it always returns @code{Pmode}. You 1273would most commonly define this macro if the 1274@code{save_stack_@var{level}} patterns need to support both a 32- and a 127564-bit mode. 1276@end defmac 1277 1278@defmac STACK_SIZE_MODE 1279If defined, an expression of type @code{enum machine_mode} that 1280specifies the mode of the size increment operand of an 1281@code{allocate_stack} named pattern (@pxref{Standard Names}). 1282 1283You need not define this macro if it always returns @code{word_mode}. 1284You would most commonly define this macro if the @code{allocate_stack} 1285pattern needs to support both a 32- and a 64-bit mode. 1286@end defmac 1287 1288@defmac TARGET_FLOAT_FORMAT 1289A code distinguishing the floating point format of the target machine. 1290There are four defined values: 1291 1292@ftable @code 1293@item IEEE_FLOAT_FORMAT 1294This code indicates IEEE floating point. It is the default; there is no 1295need to define @code{TARGET_FLOAT_FORMAT} when the format is IEEE@. 1296 1297@item VAX_FLOAT_FORMAT 1298This code indicates the ``F float'' (for @code{float}) and ``D float'' 1299or ``G float'' formats (for @code{double}) used on the VAX and PDP-11@. 1300 1301@item IBM_FLOAT_FORMAT 1302This code indicates the format used on the IBM System/370. 1303 1304@item C4X_FLOAT_FORMAT 1305This code indicates the format used on the TMS320C3x/C4x. 1306@end ftable 1307 1308If your target uses a floating point format other than these, you must 1309define a new @var{name}_FLOAT_FORMAT code for it, and add support for 1310it to @file{real.c}. 1311 1312The ordering of the component words of floating point values stored in 1313memory is controlled by @code{FLOAT_WORDS_BIG_ENDIAN}. 1314@end defmac 1315 1316@defmac MODE_HAS_NANS (@var{mode}) 1317When defined, this macro should be true if @var{mode} has a NaN 1318representation. The compiler assumes that NaNs are not equal to 1319anything (including themselves) and that addition, subtraction, 1320multiplication and division all return NaNs when one operand is 1321NaN@. 1322 1323By default, this macro is true if @var{mode} is a floating-point 1324mode and the target floating-point format is IEEE@. 1325@end defmac 1326 1327@defmac MODE_HAS_INFINITIES (@var{mode}) 1328This macro should be true if @var{mode} can represent infinity. At 1329present, the compiler uses this macro to decide whether @samp{x - x} 1330is always defined. By default, the macro is true when @var{mode} 1331is a floating-point mode and the target format is IEEE@. 1332@end defmac 1333 1334@defmac MODE_HAS_SIGNED_ZEROS (@var{mode}) 1335True if @var{mode} distinguishes between positive and negative zero. 1336The rules are expected to follow the IEEE standard: 1337 1338@itemize @bullet 1339@item 1340@samp{x + x} has the same sign as @samp{x}. 1341 1342@item 1343If the sum of two values with opposite sign is zero, the result is 1344positive for all rounding modes expect towards @minus{}infinity, for 1345which it is negative. 1346 1347@item 1348The sign of a product or quotient is negative when exactly one 1349of the operands is negative. 1350@end itemize 1351 1352The default definition is true if @var{mode} is a floating-point 1353mode and the target format is IEEE@. 1354@end defmac 1355 1356@defmac MODE_HAS_SIGN_DEPENDENT_ROUNDING (@var{mode}) 1357If defined, this macro should be true for @var{mode} if it has at 1358least one rounding mode in which @samp{x} and @samp{-x} can be 1359rounded to numbers of different magnitude. Two such modes are 1360towards @minus{}infinity and towards +infinity. 1361 1362The default definition of this macro is true if @var{mode} is 1363a floating-point mode and the target format is IEEE@. 1364@end defmac 1365 1366@defmac ROUND_TOWARDS_ZERO 1367If defined, this macro should be true if the prevailing rounding 1368mode is towards zero. A true value has the following effects: 1369 1370@itemize @bullet 1371@item 1372@code{MODE_HAS_SIGN_DEPENDENT_ROUNDING} will be false for all modes. 1373 1374@item 1375@file{libgcc.a}'s floating-point emulator will round towards zero 1376rather than towards nearest. 1377 1378@item 1379The compiler's floating-point emulator will round towards zero after 1380doing arithmetic, and when converting from the internal float format to 1381the target format. 1382@end itemize 1383 1384The macro does not affect the parsing of string literals. When the 1385primary rounding mode is towards zero, library functions like 1386@code{strtod} might still round towards nearest, and the compiler's 1387parser should behave like the target's @code{strtod} where possible. 1388 1389Not defining this macro is equivalent to returning zero. 1390@end defmac 1391 1392@defmac LARGEST_EXPONENT_IS_NORMAL (@var{size}) 1393This macro should return true if floats with @var{size} 1394bits do not have a NaN or infinity representation, but use the largest 1395exponent for normal numbers instead. 1396 1397Defining this macro to true for @var{size} causes @code{MODE_HAS_NANS} 1398and @code{MODE_HAS_INFINITIES} to be false for @var{size}-bit modes. 1399It also affects the way @file{libgcc.a} and @file{real.c} emulate 1400floating-point arithmetic. 1401 1402The default definition of this macro returns false for all sizes. 1403@end defmac 1404 1405@deftypefn {Target Hook} bool TARGET_VECTOR_OPAQUE_P (tree @var{type}) 1406This target hook should return @code{true} a vector is opaque. That 1407is, if no cast is needed when copying a vector value of type 1408@var{type} into another vector lvalue of the same size. Vector opaque 1409types cannot be initialized. The default is that there are no such 1410types. 1411@end deftypefn 1412 1413@deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (tree @var{record_type}) 1414This target hook returns @code{true} if bit-fields in the given 1415@var{record_type} are to be laid out following the rules of Microsoft 1416Visual C/C++, namely: (i) a bit-field won't share the same storage 1417unit with the previous bit-field if their underlying types have 1418different sizes, and the bit-field will be aligned to the highest 1419alignment of the underlying types of itself and of the previous 1420bit-field; (ii) a zero-sized bit-field will affect the alignment of 1421the whole enclosing structure, even if it is unnamed; except that 1422(iii) a zero-sized bit-field will be disregarded unless it follows 1423another bit-field of nonzero size. If this hook returns @code{true}, 1424other macros that control bit-field layout are ignored. 1425 1426When a bit-field is inserted into a packed record, the whole size 1427of the underlying type is used by one or more same-size adjacent 1428bit-fields (that is, if its long:3, 32 bits is used in the record, 1429and any additional adjacent long bit-fields are packed into the same 1430chunk of 32 bits. However, if the size changes, a new field of that 1431size is allocated). In an unpacked record, this is the same as using 1432alignment, but not equivalent when packing. 1433 1434If both MS bit-fields and @samp{__attribute__((packed))} are used, 1435the latter will take precedence. If @samp{__attribute__((packed))} is 1436used on a single field when MS bit-fields are in use, it will take 1437precedence for that field, but the alignment of the rest of the structure 1438may affect its placement. 1439@end deftypefn 1440 1441@deftypefn {Target Hook} {bool} TARGET_DECIMAL_FLOAT_SUPPORTED_P (void) 1442Returns true if the target supports decimal floating point. 1443@end deftypefn 1444 1445@deftypefn {Target Hook} {const char *} TARGET_MANGLE_FUNDAMENTAL_TYPE (tree @var{type}) 1446If your target defines any fundamental types, define this hook to 1447return the appropriate encoding for these types as part of a C++ 1448mangled name. The @var{type} argument is the tree structure 1449representing the type to be mangled. The hook may be applied to trees 1450which are not target-specific fundamental types; it should return 1451@code{NULL} for all such types, as well as arguments it does not 1452recognize. If the return value is not @code{NULL}, it must point to 1453a statically-allocated string constant. 1454 1455Target-specific fundamental types might be new fundamental types or 1456qualified versions of ordinary fundamental types. Encode new 1457fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name} 1458is the name used for the type in source code, and @var{n} is the 1459length of @var{name} in decimal. Encode qualified versions of 1460ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where 1461@var{name} is the name used for the type qualifier in source code, 1462@var{n} is the length of @var{name} as above, and @var{code} is the 1463code used to represent the unqualified version of this type. (See 1464@code{write_builtin_type} in @file{cp/mangle.c} for the list of 1465codes.) In both cases the spaces are for clarity; do not include any 1466spaces in your string. 1467 1468The default version of this hook always returns @code{NULL}, which is 1469appropriate for a target that does not define any new fundamental 1470types. 1471@end deftypefn 1472 1473@node Type Layout 1474@section Layout of Source Language Data Types 1475 1476These macros define the sizes and other characteristics of the standard 1477basic data types used in programs being compiled. Unlike the macros in 1478the previous section, these apply to specific features of C and related 1479languages, rather than to fundamental aspects of storage layout. 1480 1481@defmac INT_TYPE_SIZE 1482A C expression for the size in bits of the type @code{int} on the 1483target machine. If you don't define this, the default is one word. 1484@end defmac 1485 1486@defmac SHORT_TYPE_SIZE 1487A C expression for the size in bits of the type @code{short} on the 1488target machine. If you don't define this, the default is half a word. 1489(If this would be less than one storage unit, it is rounded up to one 1490unit.) 1491@end defmac 1492 1493@defmac LONG_TYPE_SIZE 1494A C expression for the size in bits of the type @code{long} on the 1495target machine. If you don't define this, the default is one word. 1496@end defmac 1497 1498@defmac ADA_LONG_TYPE_SIZE 1499On some machines, the size used for the Ada equivalent of the type 1500@code{long} by a native Ada compiler differs from that used by C@. In 1501that situation, define this macro to be a C expression to be used for 1502the size of that type. If you don't define this, the default is the 1503value of @code{LONG_TYPE_SIZE}. 1504@end defmac 1505 1506@defmac LONG_LONG_TYPE_SIZE 1507A C expression for the size in bits of the type @code{long long} on the 1508target machine. If you don't define this, the default is two 1509words. If you want to support GNU Ada on your machine, the value of this 1510macro must be at least 64. 1511@end defmac 1512 1513@defmac CHAR_TYPE_SIZE 1514A C expression for the size in bits of the type @code{char} on the 1515target machine. If you don't define this, the default is 1516@code{BITS_PER_UNIT}. 1517@end defmac 1518 1519@defmac BOOL_TYPE_SIZE 1520A C expression for the size in bits of the C++ type @code{bool} and 1521C99 type @code{_Bool} on the target machine. If you don't define 1522this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}. 1523@end defmac 1524 1525@defmac FLOAT_TYPE_SIZE 1526A C expression for the size in bits of the type @code{float} on the 1527target machine. If you don't define this, the default is one word. 1528@end defmac 1529 1530@defmac DOUBLE_TYPE_SIZE 1531A C expression for the size in bits of the type @code{double} on the 1532target machine. If you don't define this, the default is two 1533words. 1534@end defmac 1535 1536@defmac LONG_DOUBLE_TYPE_SIZE 1537A C expression for the size in bits of the type @code{long double} on 1538the target machine. If you don't define this, the default is two 1539words. 1540@end defmac 1541 1542@defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE 1543Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or 1544if you want routines in @file{libgcc2.a} for a size other than 1545@code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the 1546default is @code{LONG_DOUBLE_TYPE_SIZE}. 1547@end defmac 1548 1549@defmac LIBGCC2_HAS_DF_MODE 1550Define this macro if neither @code{LIBGCC2_DOUBLE_TYPE_SIZE} nor 1551@code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 1552@code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a} 1553anyway. If you don't define this and either @code{LIBGCC2_DOUBLE_TYPE_SIZE} 1554or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1, 1555otherwise it is 0. 1556@end defmac 1557 1558@defmac LIBGCC2_HAS_XF_MODE 1559Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not 1560@code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a} 1561anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} 1562is 80 then the default is 1, otherwise it is 0. 1563@end defmac 1564 1565@defmac LIBGCC2_HAS_TF_MODE 1566Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not 1567@code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a} 1568anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} 1569is 128 then the default is 1, otherwise it is 0. 1570@end defmac 1571 1572@defmac SF_SIZE 1573@defmacx DF_SIZE 1574@defmacx XF_SIZE 1575@defmacx TF_SIZE 1576Define these macros to be the size in bits of the mantissa of 1577@code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values, 1578if the defaults in @file{libgcc2.h} are inappropriate. By default, 1579@code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG} 1580for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or 1581@code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether 1582@code{LIBGCC2_DOUBLE_TYPE_SIZE} or 1583@code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64. 1584@end defmac 1585 1586@defmac TARGET_FLT_EVAL_METHOD 1587A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h}, 1588assuming, if applicable, that the floating-point control word is in its 1589default state. If you do not define this macro the value of 1590@code{FLT_EVAL_METHOD} will be zero. 1591@end defmac 1592 1593@defmac WIDEST_HARDWARE_FP_SIZE 1594A C expression for the size in bits of the widest floating-point format 1595supported by the hardware. If you define this macro, you must specify a 1596value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}. 1597If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE} 1598is the default. 1599@end defmac 1600 1601@defmac DEFAULT_SIGNED_CHAR 1602An expression whose value is 1 or 0, according to whether the type 1603@code{char} should be signed or unsigned by default. The user can 1604always override this default with the options @option{-fsigned-char} 1605and @option{-funsigned-char}. 1606@end defmac 1607 1608@deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void) 1609This target hook should return true if the compiler should give an 1610@code{enum} type only as many bytes as it takes to represent the range 1611of possible values of that type. It should return false if all 1612@code{enum} types should be allocated like @code{int}. 1613 1614The default is to return false. 1615@end deftypefn 1616 1617@defmac SIZE_TYPE 1618A C expression for a string describing the name of the data type to use 1619for size values. The typedef name @code{size_t} is defined using the 1620contents of the string. 1621 1622The string can contain more than one keyword. If so, separate them with 1623spaces, and write first any length keyword, then @code{unsigned} if 1624appropriate, and finally @code{int}. The string must exactly match one 1625of the data type names defined in the function 1626@code{init_decl_processing} in the file @file{c-decl.c}. You may not 1627omit @code{int} or change the order---that would cause the compiler to 1628crash on startup. 1629 1630If you don't define this macro, the default is @code{"long unsigned 1631int"}. 1632@end defmac 1633 1634@defmac PTRDIFF_TYPE 1635A C expression for a string describing the name of the data type to use 1636for the result of subtracting two pointers. The typedef name 1637@code{ptrdiff_t} is defined using the contents of the string. See 1638@code{SIZE_TYPE} above for more information. 1639 1640If you don't define this macro, the default is @code{"long int"}. 1641@end defmac 1642 1643@defmac WCHAR_TYPE 1644A C expression for a string describing the name of the data type to use 1645for wide characters. The typedef name @code{wchar_t} is defined using 1646the contents of the string. See @code{SIZE_TYPE} above for more 1647information. 1648 1649If you don't define this macro, the default is @code{"int"}. 1650@end defmac 1651 1652@defmac WCHAR_TYPE_SIZE 1653A C expression for the size in bits of the data type for wide 1654characters. This is used in @code{cpp}, which cannot make use of 1655@code{WCHAR_TYPE}. 1656@end defmac 1657 1658@defmac WINT_TYPE 1659A C expression for a string describing the name of the data type to 1660use for wide characters passed to @code{printf} and returned from 1661@code{getwc}. The typedef name @code{wint_t} is defined using the 1662contents of the string. See @code{SIZE_TYPE} above for more 1663information. 1664 1665If you don't define this macro, the default is @code{"unsigned int"}. 1666@end defmac 1667 1668@defmac INTMAX_TYPE 1669A C expression for a string describing the name of the data type that 1670can represent any value of any standard or extended signed integer type. 1671The typedef name @code{intmax_t} is defined using the contents of the 1672string. See @code{SIZE_TYPE} above for more information. 1673 1674If you don't define this macro, the default is the first of 1675@code{"int"}, @code{"long int"}, or @code{"long long int"} that has as 1676much precision as @code{long long int}. 1677@end defmac 1678 1679@defmac UINTMAX_TYPE 1680A C expression for a string describing the name of the data type that 1681can represent any value of any standard or extended unsigned integer 1682type. The typedef name @code{uintmax_t} is defined using the contents 1683of the string. See @code{SIZE_TYPE} above for more information. 1684 1685If you don't define this macro, the default is the first of 1686@code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long 1687unsigned int"} that has as much precision as @code{long long unsigned 1688int}. 1689@end defmac 1690 1691@defmac TARGET_PTRMEMFUNC_VBIT_LOCATION 1692The C++ compiler represents a pointer-to-member-function with a struct 1693that looks like: 1694 1695@smallexample 1696 struct @{ 1697 union @{ 1698 void (*fn)(); 1699 ptrdiff_t vtable_index; 1700 @}; 1701 ptrdiff_t delta; 1702 @}; 1703@end smallexample 1704 1705@noindent 1706The C++ compiler must use one bit to indicate whether the function that 1707will be called through a pointer-to-member-function is virtual. 1708Normally, we assume that the low-order bit of a function pointer must 1709always be zero. Then, by ensuring that the vtable_index is odd, we can 1710distinguish which variant of the union is in use. But, on some 1711platforms function pointers can be odd, and so this doesn't work. In 1712that case, we use the low-order bit of the @code{delta} field, and shift 1713the remainder of the @code{delta} field to the left. 1714 1715GCC will automatically make the right selection about where to store 1716this bit using the @code{FUNCTION_BOUNDARY} setting for your platform. 1717However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY} 1718set such that functions always start at even addresses, but the lowest 1719bit of pointers to functions indicate whether the function at that 1720address is in ARM or Thumb mode. If this is the case of your 1721architecture, you should define this macro to 1722@code{ptrmemfunc_vbit_in_delta}. 1723 1724In general, you should not have to define this macro. On architectures 1725in which function addresses are always even, according to 1726@code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to 1727@code{ptrmemfunc_vbit_in_pfn}. 1728@end defmac 1729 1730@defmac TARGET_VTABLE_USES_DESCRIPTORS 1731Normally, the C++ compiler uses function pointers in vtables. This 1732macro allows the target to change to use ``function descriptors'' 1733instead. Function descriptors are found on targets for whom a 1734function pointer is actually a small data structure. Normally the 1735data structure consists of the actual code address plus a data 1736pointer to which the function's data is relative. 1737 1738If vtables are used, the value of this macro should be the number 1739of words that the function descriptor occupies. 1740@end defmac 1741 1742@defmac TARGET_VTABLE_ENTRY_ALIGN 1743By default, the vtable entries are void pointers, the so the alignment 1744is the same as pointer alignment. The value of this macro specifies 1745the alignment of the vtable entry in bits. It should be defined only 1746when special alignment is necessary. */ 1747@end defmac 1748 1749@defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE 1750There are a few non-descriptor entries in the vtable at offsets below 1751zero. If these entries must be padded (say, to preserve the alignment 1752specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number 1753of words in each data entry. 1754@end defmac 1755 1756@node Registers 1757@section Register Usage 1758@cindex register usage 1759 1760This section explains how to describe what registers the target machine 1761has, and how (in general) they can be used. 1762 1763The description of which registers a specific instruction can use is 1764done with register classes; see @ref{Register Classes}. For information 1765on using registers to access a stack frame, see @ref{Frame Registers}. 1766For passing values in registers, see @ref{Register Arguments}. 1767For returning values in registers, see @ref{Scalar Return}. 1768 1769@menu 1770* Register Basics:: Number and kinds of registers. 1771* Allocation Order:: Order in which registers are allocated. 1772* Values in Registers:: What kinds of values each reg can hold. 1773* Leaf Functions:: Renumbering registers for leaf functions. 1774* Stack Registers:: Handling a register stack such as 80387. 1775@end menu 1776 1777@node Register Basics 1778@subsection Basic Characteristics of Registers 1779 1780@c prevent bad page break with this line 1781Registers have various characteristics. 1782 1783@defmac FIRST_PSEUDO_REGISTER 1784Number of hardware registers known to the compiler. They receive 1785numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first 1786pseudo register's number really is assigned the number 1787@code{FIRST_PSEUDO_REGISTER}. 1788@end defmac 1789 1790@defmac FIXED_REGISTERS 1791@cindex fixed register 1792An initializer that says which registers are used for fixed purposes 1793all throughout the compiled code and are therefore not available for 1794general allocation. These would include the stack pointer, the frame 1795pointer (except on machines where that can be used as a general 1796register when no frame pointer is needed), the program counter on 1797machines where that is considered one of the addressable registers, 1798and any other numbered register with a standard use. 1799 1800This information is expressed as a sequence of numbers, separated by 1801commas and surrounded by braces. The @var{n}th number is 1 if 1802register @var{n} is fixed, 0 otherwise. 1803 1804The table initialized from this macro, and the table initialized by 1805the following one, may be overridden at run time either automatically, 1806by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by 1807the user with the command options @option{-ffixed-@var{reg}}, 1808@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}. 1809@end defmac 1810 1811@defmac CALL_USED_REGISTERS 1812@cindex call-used register 1813@cindex call-clobbered register 1814@cindex call-saved register 1815Like @code{FIXED_REGISTERS} but has 1 for each register that is 1816clobbered (in general) by function calls as well as for fixed 1817registers. This macro therefore identifies the registers that are not 1818available for general allocation of values that must live across 1819function calls. 1820 1821If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler 1822automatically saves it on function entry and restores it on function 1823exit, if the register is used within the function. 1824@end defmac 1825 1826@defmac CALL_REALLY_USED_REGISTERS 1827@cindex call-used register 1828@cindex call-clobbered register 1829@cindex call-saved register 1830Like @code{CALL_USED_REGISTERS} except this macro doesn't require 1831that the entire set of @code{FIXED_REGISTERS} be included. 1832(@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}). 1833This macro is optional. If not specified, it defaults to the value 1834of @code{CALL_USED_REGISTERS}. 1835@end defmac 1836 1837@defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode}) 1838@cindex call-used register 1839@cindex call-clobbered register 1840@cindex call-saved register 1841A C expression that is nonzero if it is not permissible to store a 1842value of mode @var{mode} in hard register number @var{regno} across a 1843call without some part of it being clobbered. For most machines this 1844macro need not be defined. It is only required for machines that do not 1845preserve the entire contents of a register across a call. 1846@end defmac 1847 1848@findex fixed_regs 1849@findex call_used_regs 1850@findex global_regs 1851@findex reg_names 1852@findex reg_class_contents 1853@defmac CONDITIONAL_REGISTER_USAGE 1854Zero or more C statements that may conditionally modify five variables 1855@code{fixed_regs}, @code{call_used_regs}, @code{global_regs}, 1856@code{reg_names}, and @code{reg_class_contents}, to take into account 1857any dependence of these register sets on target flags. The first three 1858of these are of type @code{char []} (interpreted as Boolean vectors). 1859@code{global_regs} is a @code{const char *[]}, and 1860@code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is 1861called, @code{fixed_regs}, @code{call_used_regs}, 1862@code{reg_class_contents}, and @code{reg_names} have been initialized 1863from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS}, 1864@code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively. 1865@code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}}, 1866@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}} 1867command options have been applied. 1868 1869You need not define this macro if it has no work to do. 1870 1871@cindex disabling certain registers 1872@cindex controlling register usage 1873If the usage of an entire class of registers depends on the target 1874flags, you may indicate this to GCC by using this macro to modify 1875@code{fixed_regs} and @code{call_used_regs} to 1 for each of the 1876registers in the classes which should not be used by GCC@. Also define 1877the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT} 1878to return @code{NO_REGS} if it 1879is called with a letter for a class that shouldn't be used. 1880 1881(However, if this class is not included in @code{GENERAL_REGS} and all 1882of the insn patterns whose constraints permit this class are 1883controlled by target switches, then GCC will automatically avoid using 1884these registers when the target switches are opposed to them.) 1885@end defmac 1886 1887@defmac INCOMING_REGNO (@var{out}) 1888Define this macro if the target machine has register windows. This C 1889expression returns the register number as seen by the called function 1890corresponding to the register number @var{out} as seen by the calling 1891function. Return @var{out} if register number @var{out} is not an 1892outbound register. 1893@end defmac 1894 1895@defmac OUTGOING_REGNO (@var{in}) 1896Define this macro if the target machine has register windows. This C 1897expression returns the register number as seen by the calling function 1898corresponding to the register number @var{in} as seen by the called 1899function. Return @var{in} if register number @var{in} is not an inbound 1900register. 1901@end defmac 1902 1903@defmac LOCAL_REGNO (@var{regno}) 1904Define this macro if the target machine has register windows. This C 1905expression returns true if the register is call-saved but is in the 1906register window. Unlike most call-saved registers, such registers 1907need not be explicitly restored on function exit or during non-local 1908gotos. 1909@end defmac 1910 1911@defmac PC_REGNUM 1912If the program counter has a register number, define this as that 1913register number. Otherwise, do not define it. 1914@end defmac 1915 1916@node Allocation Order 1917@subsection Order of Allocation of Registers 1918@cindex order of register allocation 1919@cindex register allocation order 1920 1921@c prevent bad page break with this line 1922Registers are allocated in order. 1923 1924@defmac REG_ALLOC_ORDER 1925If defined, an initializer for a vector of integers, containing the 1926numbers of hard registers in the order in which GCC should prefer 1927to use them (from most preferred to least). 1928 1929If this macro is not defined, registers are used lowest numbered first 1930(all else being equal). 1931 1932One use of this macro is on machines where the highest numbered 1933registers must always be saved and the save-multiple-registers 1934instruction supports only sequences of consecutive registers. On such 1935machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists 1936the highest numbered allocable register first. 1937@end defmac 1938 1939@defmac ORDER_REGS_FOR_LOCAL_ALLOC 1940A C statement (sans semicolon) to choose the order in which to allocate 1941hard registers for pseudo-registers local to a basic block. 1942 1943Store the desired register order in the array @code{reg_alloc_order}. 1944Element 0 should be the register to allocate first; element 1, the next 1945register; and so on. 1946 1947The macro body should not assume anything about the contents of 1948@code{reg_alloc_order} before execution of the macro. 1949 1950On most machines, it is not necessary to define this macro. 1951@end defmac 1952 1953@node Values in Registers 1954@subsection How Values Fit in Registers 1955 1956This section discusses the macros that describe which kinds of values 1957(specifically, which machine modes) each register can hold, and how many 1958consecutive registers are needed for a given mode. 1959 1960@defmac HARD_REGNO_NREGS (@var{regno}, @var{mode}) 1961A C expression for the number of consecutive hard registers, starting 1962at register number @var{regno}, required to hold a value of mode 1963@var{mode}. 1964 1965On a machine where all registers are exactly one word, a suitable 1966definition of this macro is 1967 1968@smallexample 1969#define HARD_REGNO_NREGS(REGNO, MODE) \ 1970 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \ 1971 / UNITS_PER_WORD) 1972@end smallexample 1973@end defmac 1974 1975@defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode}) 1976A C expression that is nonzero if a value of mode @var{mode}, stored 1977in memory, ends with padding that causes it to take up more space than 1978in registers starting at register number @var{regno} (as determined by 1979multiplying GCC's notion of the size of the register when containing 1980this mode by the number of registers returned by 1981@code{HARD_REGNO_NREGS}). By default this is zero. 1982 1983For example, if a floating-point value is stored in three 32-bit 1984registers but takes up 128 bits in memory, then this would be 1985nonzero. 1986 1987This macros only needs to be defined if there are cases where 1988@code{subreg_regno_offset} and @code{subreg_offset_representable_p} 1989would otherwise wrongly determine that a @code{subreg} can be 1990represented by an offset to the register number, when in fact such a 1991@code{subreg} would contain some of the padding not stored in 1992registers and so not be representable. 1993@end defmac 1994 1995@defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode}) 1996For values of @var{regno} and @var{mode} for which 1997@code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression 1998returning the greater number of registers required to hold the value 1999including any padding. In the example above, the value would be four. 2000@end defmac 2001 2002@defmac REGMODE_NATURAL_SIZE (@var{mode}) 2003Define this macro if the natural size of registers that hold values 2004of mode @var{mode} is not the word size. It is a C expression that 2005should give the natural size in bytes for the specified mode. It is 2006used by the register allocator to try to optimize its results. This 2007happens for example on SPARC 64-bit where the natural size of 2008floating-point registers is still 32-bit. 2009@end defmac 2010 2011@defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode}) 2012A C expression that is nonzero if it is permissible to store a value 2013of mode @var{mode} in hard register number @var{regno} (or in several 2014registers starting with that one). For a machine where all registers 2015are equivalent, a suitable definition is 2016 2017@smallexample 2018#define HARD_REGNO_MODE_OK(REGNO, MODE) 1 2019@end smallexample 2020 2021You need not include code to check for the numbers of fixed registers, 2022because the allocation mechanism considers them to be always occupied. 2023 2024@cindex register pairs 2025On some machines, double-precision values must be kept in even/odd 2026register pairs. You can implement that by defining this macro to reject 2027odd register numbers for such modes. 2028 2029The minimum requirement for a mode to be OK in a register is that the 2030@samp{mov@var{mode}} instruction pattern support moves between the 2031register and other hard register in the same class and that moving a 2032value into the register and back out not alter it. 2033 2034Since the same instruction used to move @code{word_mode} will work for 2035all narrower integer modes, it is not necessary on any machine for 2036@code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided 2037you define patterns @samp{movhi}, etc., to take advantage of this. This 2038is useful because of the interaction between @code{HARD_REGNO_MODE_OK} 2039and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes 2040to be tieable. 2041 2042Many machines have special registers for floating point arithmetic. 2043Often people assume that floating point machine modes are allowed only 2044in floating point registers. This is not true. Any registers that 2045can hold integers can safely @emph{hold} a floating point machine 2046mode, whether or not floating arithmetic can be done on it in those 2047registers. Integer move instructions can be used to move the values. 2048 2049On some machines, though, the converse is true: fixed-point machine 2050modes may not go in floating registers. This is true if the floating 2051registers normalize any value stored in them, because storing a 2052non-floating value there would garble it. In this case, 2053@code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in 2054floating registers. But if the floating registers do not automatically 2055normalize, if you can store any bit pattern in one and retrieve it 2056unchanged without a trap, then any machine mode may go in a floating 2057register, so you can define this macro to say so. 2058 2059The primary significance of special floating registers is rather that 2060they are the registers acceptable in floating point arithmetic 2061instructions. However, this is of no concern to 2062@code{HARD_REGNO_MODE_OK}. You handle it by writing the proper 2063constraints for those instructions. 2064 2065On some machines, the floating registers are especially slow to access, 2066so that it is better to store a value in a stack frame than in such a 2067register if floating point arithmetic is not being done. As long as the 2068floating registers are not in class @code{GENERAL_REGS}, they will not 2069be used unless some pattern's constraint asks for one. 2070@end defmac 2071 2072@defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to}) 2073A C expression that is nonzero if it is OK to rename a hard register 2074@var{from} to another hard register @var{to}. 2075 2076One common use of this macro is to prevent renaming of a register to 2077another register that is not saved by a prologue in an interrupt 2078handler. 2079 2080The default is always nonzero. 2081@end defmac 2082 2083@defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2}) 2084A C expression that is nonzero if a value of mode 2085@var{mode1} is accessible in mode @var{mode2} without copying. 2086 2087If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and 2088@code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for 2089any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})} 2090should be nonzero. If they differ for any @var{r}, you should define 2091this macro to return zero unless some other mechanism ensures the 2092accessibility of the value in a narrower mode. 2093 2094You should define this macro to return nonzero in as many cases as 2095possible since doing so will allow GCC to perform better register 2096allocation. 2097@end defmac 2098 2099@defmac AVOID_CCMODE_COPIES 2100Define this macro if the compiler should avoid copies to/from @code{CCmode} 2101registers. You should only define this macro if support for copying to/from 2102@code{CCmode} is incomplete. 2103@end defmac 2104 2105@node Leaf Functions 2106@subsection Handling Leaf Functions 2107 2108@cindex leaf functions 2109@cindex functions, leaf 2110On some machines, a leaf function (i.e., one which makes no calls) can run 2111more efficiently if it does not make its own register window. Often this 2112means it is required to receive its arguments in the registers where they 2113are passed by the caller, instead of the registers where they would 2114normally arrive. 2115 2116The special treatment for leaf functions generally applies only when 2117other conditions are met; for example, often they may use only those 2118registers for its own variables and temporaries. We use the term ``leaf 2119function'' to mean a function that is suitable for this special 2120handling, so that functions with no calls are not necessarily ``leaf 2121functions''. 2122 2123GCC assigns register numbers before it knows whether the function is 2124suitable for leaf function treatment. So it needs to renumber the 2125registers in order to output a leaf function. The following macros 2126accomplish this. 2127 2128@defmac LEAF_REGISTERS 2129Name of a char vector, indexed by hard register number, which 2130contains 1 for a register that is allowable in a candidate for leaf 2131function treatment. 2132 2133If leaf function treatment involves renumbering the registers, then the 2134registers marked here should be the ones before renumbering---those that 2135GCC would ordinarily allocate. The registers which will actually be 2136used in the assembler code, after renumbering, should not be marked with 1 2137in this vector. 2138 2139Define this macro only if the target machine offers a way to optimize 2140the treatment of leaf functions. 2141@end defmac 2142 2143@defmac LEAF_REG_REMAP (@var{regno}) 2144A C expression whose value is the register number to which @var{regno} 2145should be renumbered, when a function is treated as a leaf function. 2146 2147If @var{regno} is a register number which should not appear in a leaf 2148function before renumbering, then the expression should yield @minus{}1, which 2149will cause the compiler to abort. 2150 2151Define this macro only if the target machine offers a way to optimize the 2152treatment of leaf functions, and registers need to be renumbered to do 2153this. 2154@end defmac 2155 2156@findex current_function_is_leaf 2157@findex current_function_uses_only_leaf_regs 2158@code{TARGET_ASM_FUNCTION_PROLOGUE} and 2159@code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions 2160specially. They can test the C variable @code{current_function_is_leaf} 2161which is nonzero for leaf functions. @code{current_function_is_leaf} is 2162set prior to local register allocation and is valid for the remaining 2163compiler passes. They can also test the C variable 2164@code{current_function_uses_only_leaf_regs} which is nonzero for leaf 2165functions which only use leaf registers. 2166@code{current_function_uses_only_leaf_regs} is valid after all passes 2167that modify the instructions have been run and is only useful if 2168@code{LEAF_REGISTERS} is defined. 2169@c changed this to fix overfull. ALSO: why the "it" at the beginning 2170@c of the next paragraph?! --mew 2feb93 2171 2172@node Stack Registers 2173@subsection Registers That Form a Stack 2174 2175There are special features to handle computers where some of the 2176``registers'' form a stack. Stack registers are normally written by 2177pushing onto the stack, and are numbered relative to the top of the 2178stack. 2179 2180Currently, GCC can only handle one group of stack-like registers, and 2181they must be consecutively numbered. Furthermore, the existing 2182support for stack-like registers is specific to the 80387 floating 2183point coprocessor. If you have a new architecture that uses 2184stack-like registers, you will need to do substantial work on 2185@file{reg-stack.c} and write your machine description to cooperate 2186with it, as well as defining these macros. 2187 2188@defmac STACK_REGS 2189Define this if the machine has any stack-like registers. 2190@end defmac 2191 2192@defmac FIRST_STACK_REG 2193The number of the first stack-like register. This one is the top 2194of the stack. 2195@end defmac 2196 2197@defmac LAST_STACK_REG 2198The number of the last stack-like register. This one is the bottom of 2199the stack. 2200@end defmac 2201 2202@node Register Classes 2203@section Register Classes 2204@cindex register class definitions 2205@cindex class definitions, register 2206 2207On many machines, the numbered registers are not all equivalent. 2208For example, certain registers may not be allowed for indexed addressing; 2209certain registers may not be allowed in some instructions. These machine 2210restrictions are described to the compiler using @dfn{register classes}. 2211 2212You define a number of register classes, giving each one a name and saying 2213which of the registers belong to it. Then you can specify register classes 2214that are allowed as operands to particular instruction patterns. 2215 2216@findex ALL_REGS 2217@findex NO_REGS 2218In general, each register will belong to several classes. In fact, one 2219class must be named @code{ALL_REGS} and contain all the registers. Another 2220class must be named @code{NO_REGS} and contain no registers. Often the 2221union of two classes will be another class; however, this is not required. 2222 2223@findex GENERAL_REGS 2224One of the classes must be named @code{GENERAL_REGS}. There is nothing 2225terribly special about the name, but the operand constraint letters 2226@samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is 2227the same as @code{ALL_REGS}, just define it as a macro which expands 2228to @code{ALL_REGS}. 2229 2230Order the classes so that if class @var{x} is contained in class @var{y} 2231then @var{x} has a lower class number than @var{y}. 2232 2233The way classes other than @code{GENERAL_REGS} are specified in operand 2234constraints is through machine-dependent operand constraint letters. 2235You can define such letters to correspond to various classes, then use 2236them in operand constraints. 2237 2238You should define a class for the union of two classes whenever some 2239instruction allows both classes. For example, if an instruction allows 2240either a floating point (coprocessor) register or a general register for a 2241certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS} 2242which includes both of them. Otherwise you will get suboptimal code. 2243 2244You must also specify certain redundant information about the register 2245classes: for each class, which classes contain it and which ones are 2246contained in it; for each pair of classes, the largest class contained 2247in their union. 2248 2249When a value occupying several consecutive registers is expected in a 2250certain class, all the registers used must belong to that class. 2251Therefore, register classes cannot be used to enforce a requirement for 2252a register pair to start with an even-numbered register. The way to 2253specify this requirement is with @code{HARD_REGNO_MODE_OK}. 2254 2255Register classes used for input-operands of bitwise-and or shift 2256instructions have a special requirement: each such class must have, for 2257each fixed-point machine mode, a subclass whose registers can transfer that 2258mode to or from memory. For example, on some machines, the operations for 2259single-byte values (@code{QImode}) are limited to certain registers. When 2260this is so, each register class that is used in a bitwise-and or shift 2261instruction must have a subclass consisting of registers from which 2262single-byte values can be loaded or stored. This is so that 2263@code{PREFERRED_RELOAD_CLASS} can always have a possible value to return. 2264 2265@deftp {Data type} {enum reg_class} 2266An enumerated type that must be defined with all the register class names 2267as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS} 2268must be the last register class, followed by one more enumerated value, 2269@code{LIM_REG_CLASSES}, which is not a register class but rather 2270tells how many classes there are. 2271 2272Each register class has a number, which is the value of casting 2273the class name to type @code{int}. The number serves as an index 2274in many of the tables described below. 2275@end deftp 2276 2277@defmac N_REG_CLASSES 2278The number of distinct register classes, defined as follows: 2279 2280@smallexample 2281#define N_REG_CLASSES (int) LIM_REG_CLASSES 2282@end smallexample 2283@end defmac 2284 2285@defmac REG_CLASS_NAMES 2286An initializer containing the names of the register classes as C string 2287constants. These names are used in writing some of the debugging dumps. 2288@end defmac 2289 2290@defmac REG_CLASS_CONTENTS 2291An initializer containing the contents of the register classes, as integers 2292which are bit masks. The @var{n}th integer specifies the contents of class 2293@var{n}. The way the integer @var{mask} is interpreted is that 2294register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1. 2295 2296When the machine has more than 32 registers, an integer does not suffice. 2297Then the integers are replaced by sub-initializers, braced groupings containing 2298several integers. Each sub-initializer must be suitable as an initializer 2299for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}. 2300In this situation, the first integer in each sub-initializer corresponds to 2301registers 0 through 31, the second integer to registers 32 through 63, and 2302so on. 2303@end defmac 2304 2305@defmac REGNO_REG_CLASS (@var{regno}) 2306A C expression whose value is a register class containing hard register 2307@var{regno}. In general there is more than one such class; choose a class 2308which is @dfn{minimal}, meaning that no smaller class also contains the 2309register. 2310@end defmac 2311 2312@defmac BASE_REG_CLASS 2313A macro whose definition is the name of the class to which a valid 2314base register must belong. A base register is one used in an address 2315which is the register value plus a displacement. 2316@end defmac 2317 2318@defmac MODE_BASE_REG_CLASS (@var{mode}) 2319This is a variation of the @code{BASE_REG_CLASS} macro which allows 2320the selection of a base register in a mode dependent manner. If 2321@var{mode} is VOIDmode then it should return the same value as 2322@code{BASE_REG_CLASS}. 2323@end defmac 2324 2325@defmac MODE_BASE_REG_REG_CLASS (@var{mode}) 2326A C expression whose value is the register class to which a valid 2327base register must belong in order to be used in a base plus index 2328register address. You should define this macro if base plus index 2329addresses have different requirements than other base register uses. 2330@end defmac 2331 2332@defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{outer_code}, @var{index_code}) 2333A C expression whose value is the register class to which a valid 2334base register must belong. @var{outer_code} and @var{index_code} define the 2335context in which the base register occurs. @var{outer_code} is the code of 2336the immediately enclosing expression (@code{MEM} for the top level of an 2337address, @code{ADDRESS} for something that occurs in an 2338@code{address_operand}). @var{index_code} is the code of the corresponding 2339index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise. 2340@end defmac 2341 2342@defmac INDEX_REG_CLASS 2343A macro whose definition is the name of the class to which a valid 2344index register must belong. An index register is one used in an 2345address where its value is either multiplied by a scale factor or 2346added to another register (as well as added to a displacement). 2347@end defmac 2348 2349@defmac REGNO_OK_FOR_BASE_P (@var{num}) 2350A C expression which is nonzero if register number @var{num} is 2351suitable for use as a base register in operand addresses. It may be 2352either a suitable hard register or a pseudo register that has been 2353allocated such a hard register. 2354@end defmac 2355 2356@defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode}) 2357A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that 2358that expression may examine the mode of the memory reference in 2359@var{mode}. You should define this macro if the mode of the memory 2360reference affects whether a register may be used as a base register. If 2361you define this macro, the compiler will use it instead of 2362@code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for addresses 2363that appear outside a @code{MEM}, i.e. as an @code{address_operand}. 2364 2365@end defmac 2366 2367@defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode}) 2368A C expression which is nonzero if register number @var{num} is suitable for 2369use as a base register in base plus index operand addresses, accessing 2370memory in mode @var{mode}. It may be either a suitable hard register or a 2371pseudo register that has been allocated such a hard register. You should 2372define this macro if base plus index addresses have different requirements 2373than other base register uses. 2374 2375Use of this macro is deprecated; please use the more general 2376@code{REGNO_MODE_CODE_OK_FOR_BASE_P}. 2377@end defmac 2378 2379@defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{outer_code}, @var{index_code}) 2380A C expression that is just like @code{REGNO_MODE_OK_FOR_BASE_P}, except that 2381that expression may examine the context in which the register appears in the 2382memory reference. @var{outer_code} is the code of the immediately enclosing 2383expression (@code{MEM} if at the top level of the address, @code{ADDRESS} for 2384something that occurs in an @code{address_operand}). @var{index_code} is the 2385code of the corresponding index expression if @var{outer_code} is @code{PLUS}; 2386@code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses 2387that appear outside a @code{MEM}, i.e. as an @code{address_operand}. 2388@end defmac 2389 2390@defmac REGNO_OK_FOR_INDEX_P (@var{num}) 2391A C expression which is nonzero if register number @var{num} is 2392suitable for use as an index register in operand addresses. It may be 2393either a suitable hard register or a pseudo register that has been 2394allocated such a hard register. 2395 2396The difference between an index register and a base register is that 2397the index register may be scaled. If an address involves the sum of 2398two registers, neither one of them scaled, then either one may be 2399labeled the ``base'' and the other the ``index''; but whichever 2400labeling is used must fit the machine's constraints of which registers 2401may serve in each capacity. The compiler will try both labelings, 2402looking for one that is valid, and will reload one or both registers 2403only if neither labeling works. 2404@end defmac 2405 2406@defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class}) 2407A C expression that places additional restrictions on the register class 2408to use when it is necessary to copy value @var{x} into a register in class 2409@var{class}. The value is a register class; perhaps @var{class}, or perhaps 2410another, smaller class. On many machines, the following definition is 2411safe: 2412 2413@smallexample 2414#define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS 2415@end smallexample 2416 2417Sometimes returning a more restrictive class makes better code. For 2418example, on the 68000, when @var{x} is an integer constant that is in range 2419for a @samp{moveq} instruction, the value of this macro is always 2420@code{DATA_REGS} as long as @var{class} includes the data registers. 2421Requiring a data register guarantees that a @samp{moveq} will be used. 2422 2423One case where @code{PREFERRED_RELOAD_CLASS} must not return 2424@var{class} is if @var{x} is a legitimate constant which cannot be 2425loaded into some register class. By returning @code{NO_REGS} you can 2426force @var{x} into a memory location. For example, rs6000 can load 2427immediate values into general-purpose registers, but does not have an 2428instruction for loading an immediate value into a floating-point 2429register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when 2430@var{x} is a floating-point constant. If the constant can't be loaded 2431into any kind of register, code generation will be better if 2432@code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead 2433of using @code{PREFERRED_RELOAD_CLASS}. 2434 2435If an insn has pseudos in it after register allocation, reload will go 2436through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS} 2437to find the best one. Returning @code{NO_REGS}, in this case, makes 2438reload add a @code{!} in front of the constraint: the x86 back-end uses 2439this feature to discourage usage of 387 registers when math is done in 2440the SSE registers (and vice versa). 2441@end defmac 2442 2443@defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class}) 2444Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of 2445input reloads. If you don't define this macro, the default is to use 2446@var{class}, unchanged. 2447 2448You can also use @code{PREFERRED_OUTPUT_RELOAD_CLASS} to discourage 2449reload from using some alternatives, like @code{PREFERRED_RELOAD_CLASS}. 2450@end defmac 2451 2452@defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class}) 2453A C expression that places additional restrictions on the register class 2454to use when it is necessary to be able to hold a value of mode 2455@var{mode} in a reload register for which class @var{class} would 2456ordinarily be used. 2457 2458Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when 2459there are certain modes that simply can't go in certain reload classes. 2460 2461The value is a register class; perhaps @var{class}, or perhaps another, 2462smaller class. 2463 2464Don't define this macro unless the target machine has limitations which 2465require the macro to do something nontrivial. 2466@end defmac 2467 2468@deftypefn {Target Hook} enum reg_class TARGET_SECONDARY_RELOAD (bool @var{in_p}, rtx @var{x}, enum reg_class @var{reload_class}, enum machine_mode @var{reload_mode}, secondary_reload_info *@var{sri}) 2469Many machines have some registers that cannot be copied directly to or 2470from memory or even from other types of registers. An example is the 2471@samp{MQ} register, which on most machines, can only be copied to or 2472from general registers, but not memory. Below, we shall be using the 2473term 'intermediate register' when a move operation cannot be performed 2474directly, but has to be done by copying the source into the intermediate 2475register first, and then copying the intermediate register to the 2476destination. An intermediate register always has the same mode as 2477source and destination. Since it holds the actual value being copied, 2478reload might apply optimizations to re-use an intermediate register 2479and eliding the copy from the source when it can determine that the 2480intermediate register still holds the required value. 2481 2482Another kind of secondary reload is required on some machines which 2483allow copying all registers to and from memory, but require a scratch 2484register for stores to some memory locations (e.g., those with symbolic 2485address on the RT, and those with certain symbolic address on the SPARC 2486when compiling PIC)@. Scratch registers need not have the same mode 2487as the value being copied, and usually hold a different value that 2488that being copied. Special patterns in the md file are needed to 2489describe how the copy is performed with the help of the scratch register; 2490these patterns also describe the number, register class(es) and mode(s) 2491of the scratch register(s). 2492 2493In some cases, both an intermediate and a scratch register are required. 2494 2495For input reloads, this target hook is called with nonzero @var{in_p}, 2496and @var{x} is an rtx that needs to be copied to a register in of class 2497@var{reload_class} in @var{reload_mode}. For output reloads, this target 2498hook is called with zero @var{in_p}, and a register of class @var{reload_mode} 2499needs to be copied to rtx @var{x} in @var{reload_mode}. 2500 2501If copying a register of @var{reload_class} from/to @var{x} requires 2502an intermediate register, the hook @code{secondary_reload} should 2503return the register class required for this intermediate register. 2504If no intermediate register is required, it should return NO_REGS. 2505If more than one intermediate register is required, describe the one 2506that is closest in the copy chain to the reload register. 2507 2508If scratch registers are needed, you also have to describe how to 2509perform the copy from/to the reload register to/from this 2510closest intermediate register. Or if no intermediate register is 2511required, but still a scratch register is needed, describe the 2512copy from/to the reload register to/from the reload operand @var{x}. 2513 2514You do this by setting @code{sri->icode} to the instruction code of a pattern 2515in the md file which performs the move. Operands 0 and 1 are the output 2516and input of this copy, respectively. Operands from operand 2 onward are 2517for scratch operands. These scratch operands must have a mode, and a 2518single-register-class 2519@c [later: or memory] 2520output constraint. 2521 2522When an intermediate register is used, the @code{secondary_reload} 2523hook will be called again to determine how to copy the intermediate 2524register to/from the reload operand @var{x}, so your hook must also 2525have code to handle the register class of the intermediate operand. 2526 2527@c [For later: maybe we'll allow multi-alternative reload patterns - 2528@c the port maintainer could name a mov<mode> pattern that has clobbers - 2529@c and match the constraints of input and output to determine the required 2530@c alternative. A restriction would be that constraints used to match 2531@c against reloads registers would have to be written as register class 2532@c constraints, or we need a new target macro / hook that tells us if an 2533@c arbitrary constraint can match an unknown register of a given class. 2534@c Such a macro / hook would also be useful in other places.] 2535 2536 2537@var{x} might be a pseudo-register or a @code{subreg} of a 2538pseudo-register, which could either be in a hard register or in memory. 2539Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2540in memory and the hard register number if it is in a register. 2541 2542Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are 2543currently not supported. For the time being, you will have to continue 2544to use @code{SECONDARY_MEMORY_NEEDED} for that purpose. 2545 2546@code{copy_cost} also uses this target hook to find out how values are 2547copied. If you want it to include some extra cost for the need to allocate 2548(a) scratch register(s), set @code{sri->extra_cost} to the additional cost. 2549Or if two dependent moves are supposed to have a lower cost than the sum 2550of the individual moves due to expected fortuitous scheduling and/or special 2551forwarding logic, you can set @code{sri->extra_cost} to a negative amount. 2552@end deftypefn 2553 2554@defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2555@defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2556@defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2557These macros are obsolete, new ports should use the target hook 2558@code{TARGET_SECONDARY_RELOAD} instead. 2559 2560These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD} 2561target hook. Older ports still define these macros to indicate to the 2562reload phase that it may 2563need to allocate at least one register for a reload in addition to the 2564register to contain the data. Specifically, if copying @var{x} to a 2565register @var{class} in @var{mode} requires an intermediate register, 2566you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the 2567largest register class all of whose registers can be used as 2568intermediate registers or scratch registers. 2569 2570If copying a register @var{class} in @var{mode} to @var{x} requires an 2571intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS} 2572was supposed to be defined be defined to return the largest register 2573class required. If the 2574requirements for input and output reloads were the same, the macro 2575@code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both 2576macros identically. 2577 2578The values returned by these macros are often @code{GENERAL_REGS}. 2579Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x} 2580can be directly copied to or from a register of @var{class} in 2581@var{mode} without requiring a scratch register. Do not define this 2582macro if it would always return @code{NO_REGS}. 2583 2584If a scratch register is required (either with or without an 2585intermediate register), you were supposed to define patterns for 2586@samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required 2587(@pxref{Standard Names}. These patterns, which were normally 2588implemented with a @code{define_expand}, should be similar to the 2589@samp{mov@var{m}} patterns, except that operand 2 is the scratch 2590register. 2591 2592These patterns need constraints for the reload register and scratch 2593register that 2594contain a single register class. If the original reload register (whose 2595class is @var{class}) can meet the constraint given in the pattern, the 2596value returned by these macros is used for the class of the scratch 2597register. Otherwise, two additional reload registers are required. 2598Their classes are obtained from the constraints in the insn pattern. 2599 2600@var{x} might be a pseudo-register or a @code{subreg} of a 2601pseudo-register, which could either be in a hard register or in memory. 2602Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2603in memory and the hard register number if it is in a register. 2604 2605These macros should not be used in the case where a particular class of 2606registers can only be copied to memory and not to another class of 2607registers. In that case, secondary reload registers are not needed and 2608would not be helpful. Instead, a stack location must be used to perform 2609the copy and the @code{mov@var{m}} pattern should use memory as an 2610intermediate storage. This case often occurs between floating-point and 2611general registers. 2612@end defmac 2613 2614@defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m}) 2615Certain machines have the property that some registers cannot be copied 2616to some other registers without using memory. Define this macro on 2617those machines to be a C expression that is nonzero if objects of mode 2618@var{m} in registers of @var{class1} can only be copied to registers of 2619class @var{class2} by storing a register of @var{class1} into memory 2620and loading that memory location into a register of @var{class2}. 2621 2622Do not define this macro if its value would always be zero. 2623@end defmac 2624 2625@defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode}) 2626Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler 2627allocates a stack slot for a memory location needed for register copies. 2628If this macro is defined, the compiler instead uses the memory location 2629defined by this macro. 2630 2631Do not define this macro if you do not define 2632@code{SECONDARY_MEMORY_NEEDED}. 2633@end defmac 2634 2635@defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode}) 2636When the compiler needs a secondary memory location to copy between two 2637registers of mode @var{mode}, it normally allocates sufficient memory to 2638hold a quantity of @code{BITS_PER_WORD} bits and performs the store and 2639load operations in a mode that many bits wide and whose class is the 2640same as that of @var{mode}. 2641 2642This is right thing to do on most machines because it ensures that all 2643bits of the register are copied and prevents accesses to the registers 2644in a narrower mode, which some machines prohibit for floating-point 2645registers. 2646 2647However, this default behavior is not correct on some machines, such as 2648the DEC Alpha, that store short integers in floating-point registers 2649differently than in integer registers. On those machines, the default 2650widening will not work correctly and you must define this macro to 2651suppress that widening in some cases. See the file @file{alpha.h} for 2652details. 2653 2654Do not define this macro if you do not define 2655@code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that 2656is @code{BITS_PER_WORD} bits wide is correct for your machine. 2657@end defmac 2658 2659@defmac SMALL_REGISTER_CLASSES 2660On some machines, it is risky to let hard registers live across arbitrary 2661insns. Typically, these machines have instructions that require values 2662to be in specific registers (like an accumulator), and reload will fail 2663if the required hard register is used for another purpose across such an 2664insn. 2665 2666Define @code{SMALL_REGISTER_CLASSES} to be an expression with a nonzero 2667value on these machines. When this macro has a nonzero value, the 2668compiler will try to minimize the lifetime of hard registers. 2669 2670It is always safe to define this macro with a nonzero value, but if you 2671unnecessarily define it, you will reduce the amount of optimizations 2672that can be performed in some cases. If you do not define this macro 2673with a nonzero value when it is required, the compiler will run out of 2674spill registers and print a fatal error message. For most machines, you 2675should not define this macro at all. 2676@end defmac 2677 2678@defmac CLASS_LIKELY_SPILLED_P (@var{class}) 2679A C expression whose value is nonzero if pseudos that have been assigned 2680to registers of class @var{class} would likely be spilled because 2681registers of @var{class} are needed for spill registers. 2682 2683The default value of this macro returns 1 if @var{class} has exactly one 2684register and zero otherwise. On most machines, this default should be 2685used. Only define this macro to some other expression if pseudos 2686allocated by @file{local-alloc.c} end up in memory because their hard 2687registers were needed for spill registers. If this macro returns nonzero 2688for those classes, those pseudos will only be allocated by 2689@file{global.c}, which knows how to reallocate the pseudo to another 2690register. If there would not be another register available for 2691reallocation, you should not change the definition of this macro since 2692the only effect of such a definition would be to slow down register 2693allocation. 2694@end defmac 2695 2696@defmac CLASS_MAX_NREGS (@var{class}, @var{mode}) 2697A C expression for the maximum number of consecutive registers 2698of class @var{class} needed to hold a value of mode @var{mode}. 2699 2700This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact, 2701the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})} 2702should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, 2703@var{mode})} for all @var{regno} values in the class @var{class}. 2704 2705This macro helps control the handling of multiple-word values 2706in the reload pass. 2707@end defmac 2708 2709@defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class}) 2710If defined, a C expression that returns nonzero for a @var{class} for which 2711a change from mode @var{from} to mode @var{to} is invalid. 2712 2713For the example, loading 32-bit integer or floating-point objects into 2714floating-point registers on the Alpha extends them to 64 bits. 2715Therefore loading a 64-bit object and then storing it as a 32-bit object 2716does not store the low-order 32 bits, as would be the case for a normal 2717register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS} 2718as below: 2719 2720@smallexample 2721#define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \ 2722 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \ 2723 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0) 2724@end smallexample 2725@end defmac 2726 2727@node Old Constraints 2728@section Obsolete Macros for Defining Constraints 2729@cindex defining constraints, obsolete method 2730@cindex constraints, defining, obsolete method 2731 2732Machine-specific constraints can be defined with these macros instead 2733of the machine description constructs described in @ref{Define 2734Constraints}. This mechanism is obsolete. New ports should not use 2735it; old ports should convert to the new mechanism. 2736 2737@defmac CONSTRAINT_LEN (@var{char}, @var{str}) 2738For the constraint at the start of @var{str}, which starts with the letter 2739@var{c}, return the length. This allows you to have register class / 2740constant / extra constraints that are longer than a single letter; 2741you don't need to define this macro if you can do with single-letter 2742constraints only. The definition of this macro should use 2743DEFAULT_CONSTRAINT_LEN for all the characters that you don't want 2744to handle specially. 2745There are some sanity checks in genoutput.c that check the constraint lengths 2746for the md file, so you can also use this macro to help you while you are 2747transitioning from a byzantine single-letter-constraint scheme: when you 2748return a negative length for a constraint you want to re-use, genoutput 2749will complain about every instance where it is used in the md file. 2750@end defmac 2751 2752@defmac REG_CLASS_FROM_LETTER (@var{char}) 2753A C expression which defines the machine-dependent operand constraint 2754letters for register classes. If @var{char} is such a letter, the 2755value should be the register class corresponding to it. Otherwise, 2756the value should be @code{NO_REGS}. The register letter @samp{r}, 2757corresponding to class @code{GENERAL_REGS}, will not be passed 2758to this macro; you do not need to handle it. 2759@end defmac 2760 2761@defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str}) 2762Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string 2763passed in @var{str}, so that you can use suffixes to distinguish between 2764different variants. 2765@end defmac 2766 2767@defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c}) 2768A C expression that defines the machine-dependent operand constraint 2769letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify 2770particular ranges of integer values. If @var{c} is one of those 2771letters, the expression should check that @var{value}, an integer, is in 2772the appropriate range and return 1 if so, 0 otherwise. If @var{c} is 2773not one of those letters, the value should be 0 regardless of 2774@var{value}. 2775@end defmac 2776 2777@defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str}) 2778Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint 2779string passed in @var{str}, so that you can use suffixes to distinguish 2780between different variants. 2781@end defmac 2782 2783@defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c}) 2784A C expression that defines the machine-dependent operand constraint 2785letters that specify particular ranges of @code{const_double} values 2786(@samp{G} or @samp{H}). 2787 2788If @var{c} is one of those letters, the expression should check that 2789@var{value}, an RTX of code @code{const_double}, is in the appropriate 2790range and return 1 if so, 0 otherwise. If @var{c} is not one of those 2791letters, the value should be 0 regardless of @var{value}. 2792 2793@code{const_double} is used for all floating-point constants and for 2794@code{DImode} fixed-point constants. A given letter can accept either 2795or both kinds of values. It can use @code{GET_MODE} to distinguish 2796between these kinds. 2797@end defmac 2798 2799@defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str}) 2800Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint 2801string passed in @var{str}, so that you can use suffixes to distinguish 2802between different variants. 2803@end defmac 2804 2805@defmac EXTRA_CONSTRAINT (@var{value}, @var{c}) 2806A C expression that defines the optional machine-dependent constraint 2807letters that can be used to segregate specific types of operands, usually 2808memory references, for the target machine. Any letter that is not 2809elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} / 2810@code{REG_CLASS_FROM_CONSTRAINT} 2811may be used. Normally this macro will not be defined. 2812 2813If it is required for a particular target machine, it should return 1 2814if @var{value} corresponds to the operand type represented by the 2815constraint letter @var{c}. If @var{c} is not defined as an extra 2816constraint, the value returned should be 0 regardless of @var{value}. 2817 2818For example, on the ROMP, load instructions cannot have their output 2819in r0 if the memory reference contains a symbolic address. Constraint 2820letter @samp{Q} is defined as representing a memory address that does 2821@emph{not} contain a symbolic address. An alternative is specified with 2822a @samp{Q} constraint on the input and @samp{r} on the output. The next 2823alternative specifies @samp{m} on the input and a register class that 2824does not include r0 on the output. 2825@end defmac 2826 2827@defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str}) 2828Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed 2829in @var{str}, so that you can use suffixes to distinguish between different 2830variants. 2831@end defmac 2832 2833@defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str}) 2834A C expression that defines the optional machine-dependent constraint 2835letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should 2836be treated like memory constraints by the reload pass. 2837 2838It should return 1 if the operand type represented by the constraint 2839at the start of @var{str}, the first letter of which is the letter @var{c}, 2840 comprises a subset of all memory references including 2841all those whose address is simply a base register. This allows the reload 2842pass to reload an operand, if it does not directly correspond to the operand 2843type of @var{c}, by copying its address into a base register. 2844 2845For example, on the S/390, some instructions do not accept arbitrary 2846memory references, but only those that do not make use of an index 2847register. The constraint letter @samp{Q} is defined via 2848@code{EXTRA_CONSTRAINT} as representing a memory address of this type. 2849If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT}, 2850a @samp{Q} constraint can handle any memory operand, because the 2851reload pass knows it can be reloaded by copying the memory address 2852into a base register if required. This is analogous to the way 2853a @samp{o} constraint can handle any memory operand. 2854@end defmac 2855 2856@defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str}) 2857A C expression that defines the optional machine-dependent constraint 2858letters, amongst those accepted by @code{EXTRA_CONSTRAINT} / 2859@code{EXTRA_CONSTRAINT_STR}, that should 2860be treated like address constraints by the reload pass. 2861 2862It should return 1 if the operand type represented by the constraint 2863at the start of @var{str}, which starts with the letter @var{c}, comprises 2864a subset of all memory addresses including 2865all those that consist of just a base register. This allows the reload 2866pass to reload an operand, if it does not directly correspond to the operand 2867type of @var{str}, by copying it into a base register. 2868 2869Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only 2870be used with the @code{address_operand} predicate. It is treated 2871analogously to the @samp{p} constraint. 2872@end defmac 2873 2874@node Stack and Calling 2875@section Stack Layout and Calling Conventions 2876@cindex calling conventions 2877 2878@c prevent bad page break with this line 2879This describes the stack layout and calling conventions. 2880 2881@menu 2882* Frame Layout:: 2883* Exception Handling:: 2884* Stack Checking:: 2885* Frame Registers:: 2886* Elimination:: 2887* Stack Arguments:: 2888* Register Arguments:: 2889* Scalar Return:: 2890* Aggregate Return:: 2891* Caller Saves:: 2892* Function Entry:: 2893* Profiling:: 2894* Tail Calls:: 2895* Stack Smashing Protection:: 2896@end menu 2897 2898@node Frame Layout 2899@subsection Basic Stack Layout 2900@cindex stack frame layout 2901@cindex frame layout 2902 2903@c prevent bad page break with this line 2904Here is the basic stack layout. 2905 2906@defmac STACK_GROWS_DOWNWARD 2907Define this macro if pushing a word onto the stack moves the stack 2908pointer to a smaller address. 2909 2910When we say, ``define this macro if @dots{}'', it means that the 2911compiler checks this macro only with @code{#ifdef} so the precise 2912definition used does not matter. 2913@end defmac 2914 2915@defmac STACK_PUSH_CODE 2916This macro defines the operation used when something is pushed 2917on the stack. In RTL, a push operation will be 2918@code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})} 2919 2920The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC}, 2921and @code{POST_INC}. Which of these is correct depends on 2922the stack direction and on whether the stack pointer points 2923to the last item on the stack or whether it points to the 2924space for the next item on the stack. 2925 2926The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is 2927defined, which is almost always right, and @code{PRE_INC} otherwise, 2928which is often wrong. 2929@end defmac 2930 2931@defmac FRAME_GROWS_DOWNWARD 2932Define this macro to nonzero value if the addresses of local variable slots 2933are at negative offsets from the frame pointer. 2934@end defmac 2935 2936@defmac ARGS_GROW_DOWNWARD 2937Define this macro if successive arguments to a function occupy decreasing 2938addresses on the stack. 2939@end defmac 2940 2941@defmac STARTING_FRAME_OFFSET 2942Offset from the frame pointer to the first local variable slot to be allocated. 2943 2944If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by 2945subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}. 2946Otherwise, it is found by adding the length of the first slot to the 2947value @code{STARTING_FRAME_OFFSET}. 2948@c i'm not sure if the above is still correct.. had to change it to get 2949@c rid of an overfull. --mew 2feb93 2950@end defmac 2951 2952@defmac STACK_ALIGNMENT_NEEDED 2953Define to zero to disable final alignment of the stack during reload. 2954The nonzero default for this macro is suitable for most ports. 2955 2956On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there 2957is a register save block following the local block that doesn't require 2958alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable 2959stack alignment and do it in the backend. 2960@end defmac 2961 2962@defmac STACK_POINTER_OFFSET 2963Offset from the stack pointer register to the first location at which 2964outgoing arguments are placed. If not specified, the default value of 2965zero is used. This is the proper value for most machines. 2966 2967If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 2968the first location at which outgoing arguments are placed. 2969@end defmac 2970 2971@defmac FIRST_PARM_OFFSET (@var{fundecl}) 2972Offset from the argument pointer register to the first argument's 2973address. On some machines it may depend on the data type of the 2974function. 2975 2976If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 2977the first argument's address. 2978@end defmac 2979 2980@defmac STACK_DYNAMIC_OFFSET (@var{fundecl}) 2981Offset from the stack pointer register to an item dynamically allocated 2982on the stack, e.g., by @code{alloca}. 2983 2984The default value for this macro is @code{STACK_POINTER_OFFSET} plus the 2985length of the outgoing arguments. The default is correct for most 2986machines. See @file{function.c} for details. 2987@end defmac 2988 2989@defmac INITIAL_FRAME_ADDRESS_RTX 2990A C expression whose value is RTL representing the address of the initial 2991stack frame. This address is passed to @code{RETURN_ADDR_RTX} and 2992@code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable 2993default value will be used. Define this macro in order to make frame pointer 2994elimination work in the presence of @code{__builtin_frame_address (count)} and 2995@code{__builtin_return_address (count)} for @code{count} not equal to zero. 2996@end defmac 2997 2998@defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr}) 2999A C expression whose value is RTL representing the address in a stack 3000frame where the pointer to the caller's frame is stored. Assume that 3001@var{frameaddr} is an RTL expression for the address of the stack frame 3002itself. 3003 3004If you don't define this macro, the default is to return the value 3005of @var{frameaddr}---that is, the stack frame address is also the 3006address of the stack word that points to the previous frame. 3007@end defmac 3008 3009@defmac SETUP_FRAME_ADDRESSES 3010If defined, a C expression that produces the machine-specific code to 3011setup the stack so that arbitrary frames can be accessed. For example, 3012on the SPARC, we must flush all of the register windows to the stack 3013before we can access arbitrary stack frames. You will seldom need to 3014define this macro. 3015@end defmac 3016 3017@deftypefn {Target Hook} bool TARGET_BUILTIN_SETJMP_FRAME_VALUE () 3018This target hook should return an rtx that is used to store 3019the address of the current frame into the built in @code{setjmp} buffer. 3020The default value, @code{virtual_stack_vars_rtx}, is correct for most 3021machines. One reason you may need to define this target hook is if 3022@code{hard_frame_pointer_rtx} is the appropriate value on your machine. 3023@end deftypefn 3024 3025@defmac FRAME_ADDR_RTX (@var{frameaddr}) 3026A C expression whose value is RTL representing the value of the frame 3027address for the current frame. @var{frameaddr} is the frame pointer 3028of the current frame. This is used for __builtin_frame_address. 3029You need only define this macro if the frame address is not the same 3030as the frame pointer. Most machines do not need to define it. 3031@end defmac 3032 3033@defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr}) 3034A C expression whose value is RTL representing the value of the return 3035address for the frame @var{count} steps up from the current frame, after 3036the prologue. @var{frameaddr} is the frame pointer of the @var{count} 3037frame, or the frame pointer of the @var{count} @minus{} 1 frame if 3038@code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined. 3039 3040The value of the expression must always be the correct address when 3041@var{count} is zero, but may be @code{NULL_RTX} if there is not way to 3042determine the return address of other frames. 3043@end defmac 3044 3045@defmac RETURN_ADDR_IN_PREVIOUS_FRAME 3046Define this if the return address of a particular stack frame is accessed 3047from the frame pointer of the previous stack frame. 3048@end defmac 3049 3050@defmac INCOMING_RETURN_ADDR_RTX 3051A C expression whose value is RTL representing the location of the 3052incoming return address at the beginning of any function, before the 3053prologue. This RTL is either a @code{REG}, indicating that the return 3054value is saved in @samp{REG}, or a @code{MEM} representing a location in 3055the stack. 3056 3057You only need to define this macro if you want to support call frame 3058debugging information like that provided by DWARF 2. 3059 3060If this RTL is a @code{REG}, you should also define 3061@code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}. 3062@end defmac 3063 3064@defmac DWARF_ALT_FRAME_RETURN_COLUMN 3065A C expression whose value is an integer giving a DWARF 2 column 3066number that may be used as an alternate return column. This should 3067be defined only if @code{DWARF_FRAME_RETURN_COLUMN} is set to a 3068general register, but an alternate column needs to be used for 3069signal frames. 3070@end defmac 3071 3072@defmac DWARF_ZERO_REG 3073A C expression whose value is an integer giving a DWARF 2 register 3074number that is considered to always have the value zero. This should 3075only be defined if the target has an architected zero register, and 3076someone decided it was a good idea to use that register number to 3077terminate the stack backtrace. New ports should avoid this. 3078@end defmac 3079 3080@deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index}) 3081This target hook allows the backend to emit frame-related insns that 3082contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging 3083info engine will invoke it on insns of the form 3084@smallexample 3085(set (reg) (unspec [...] UNSPEC_INDEX)) 3086@end smallexample 3087and 3088@smallexample 3089(set (reg) (unspec_volatile [...] UNSPECV_INDEX)). 3090@end smallexample 3091to let the backend emit the call frame instructions. @var{label} is 3092the CFI label attached to the insn, @var{pattern} is the pattern of 3093the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}. 3094@end deftypefn 3095 3096@defmac INCOMING_FRAME_SP_OFFSET 3097A C expression whose value is an integer giving the offset, in bytes, 3098from the value of the stack pointer register to the top of the stack 3099frame at the beginning of any function, before the prologue. The top of 3100the frame is defined to be the value of the stack pointer in the 3101previous frame, just before the call instruction. 3102 3103You only need to define this macro if you want to support call frame 3104debugging information like that provided by DWARF 2. 3105@end defmac 3106 3107@defmac ARG_POINTER_CFA_OFFSET (@var{fundecl}) 3108A C expression whose value is an integer giving the offset, in bytes, 3109from the argument pointer to the canonical frame address (cfa). The 3110final value should coincide with that calculated by 3111@code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable 3112during virtual register instantiation. 3113 3114The default value for this macro is @code{FIRST_PARM_OFFSET (fundecl)}, 3115which is correct for most machines; in general, the arguments are found 3116immediately before the stack frame. Note that this is not the case on 3117some targets that save registers into the caller's frame, such as SPARC 3118and rs6000, and so such targets need to define this macro. 3119 3120You only need to define this macro if the default is incorrect, and you 3121want to support call frame debugging information like that provided by 3122DWARF 2. 3123@end defmac 3124 3125@defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl}) 3126If defined, a C expression whose value is an integer giving the offset 3127in bytes from the frame pointer to the canonical frame address (cfa). 3128The final value should coincide with that calculated by 3129@code{INCOMING_FRAME_SP_OFFSET}. 3130 3131Normally the CFA is calculated as an offset from the argument pointer, 3132via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is 3133variable due to the ABI, this may not be possible. If this macro is 3134defined, it implies that the virtual register instantiation should be 3135based on the frame pointer instead of the argument pointer. Only one 3136of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET} 3137should be defined. 3138@end defmac 3139 3140@defmac CFA_FRAME_BASE_OFFSET (@var{fundecl}) 3141If defined, a C expression whose value is an integer giving the offset 3142in bytes from the canonical frame address (cfa) to the frame base used 3143in DWARF 2 debug information. The default is zero. A different value 3144may reduce the size of debug information on some ports. 3145@end defmac 3146 3147@node Exception Handling 3148@subsection Exception Handling Support 3149@cindex exception handling 3150 3151@defmac EH_RETURN_DATA_REGNO (@var{N}) 3152A C expression whose value is the @var{N}th register number used for 3153data by exception handlers, or @code{INVALID_REGNUM} if fewer than 3154@var{N} registers are usable. 3155 3156The exception handling library routines communicate with the exception 3157handlers via a set of agreed upon registers. Ideally these registers 3158should be call-clobbered; it is possible to use call-saved registers, 3159but may negatively impact code size. The target must support at least 31602 data registers, but should define 4 if there are enough free registers. 3161 3162You must define this macro if you want to support call frame exception 3163handling like that provided by DWARF 2. 3164@end defmac 3165 3166@defmac EH_RETURN_STACKADJ_RTX 3167A C expression whose value is RTL representing a location in which 3168to store a stack adjustment to be applied before function return. 3169This is used to unwind the stack to an exception handler's call frame. 3170It will be assigned zero on code paths that return normally. 3171 3172Typically this is a call-clobbered hard register that is otherwise 3173untouched by the epilogue, but could also be a stack slot. 3174 3175Do not define this macro if the stack pointer is saved and restored 3176by the regular prolog and epilog code in the call frame itself; in 3177this case, the exception handling library routines will update the 3178stack location to be restored in place. Otherwise, you must define 3179this macro if you want to support call frame exception handling like 3180that provided by DWARF 2. 3181@end defmac 3182 3183@defmac EH_RETURN_HANDLER_RTX 3184A C expression whose value is RTL representing a location in which 3185to store the address of an exception handler to which we should 3186return. It will not be assigned on code paths that return normally. 3187 3188Typically this is the location in the call frame at which the normal 3189return address is stored. For targets that return by popping an 3190address off the stack, this might be a memory address just below 3191the @emph{target} call frame rather than inside the current call 3192frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already 3193been assigned, so it may be used to calculate the location of the 3194target call frame. 3195 3196Some targets have more complex requirements than storing to an 3197address calculable during initial code generation. In that case 3198the @code{eh_return} instruction pattern should be used instead. 3199 3200If you want to support call frame exception handling, you must 3201define either this macro or the @code{eh_return} instruction pattern. 3202@end defmac 3203 3204@defmac RETURN_ADDR_OFFSET 3205If defined, an integer-valued C expression for which rtl will be generated 3206to add it to the exception handler address before it is searched in the 3207exception handling tables, and to subtract it again from the address before 3208using it to return to the exception handler. 3209@end defmac 3210 3211@defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global}) 3212This macro chooses the encoding of pointers embedded in the exception 3213handling sections. If at all possible, this should be defined such 3214that the exception handling section will not require dynamic relocations, 3215and so may be read-only. 3216 3217@var{code} is 0 for data, 1 for code labels, 2 for function pointers. 3218@var{global} is true if the symbol may be affected by dynamic relocations. 3219The macro should return a combination of the @code{DW_EH_PE_*} defines 3220as found in @file{dwarf2.h}. 3221 3222If this macro is not defined, pointers will not be encoded but 3223represented directly. 3224@end defmac 3225 3226@defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done}) 3227This macro allows the target to emit whatever special magic is required 3228to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}. 3229Generic code takes care of pc-relative and indirect encodings; this must 3230be defined if the target uses text-relative or data-relative encodings. 3231 3232This is a C statement that branches to @var{done} if the format was 3233handled. @var{encoding} is the format chosen, @var{size} is the number 3234of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF} 3235to be emitted. 3236@end defmac 3237 3238@defmac MD_UNWIND_SUPPORT 3239A string specifying a file to be #include'd in unwind-dw2.c. The file 3240so included typically defines @code{MD_FALLBACK_FRAME_STATE_FOR}. 3241@end defmac 3242 3243@defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs}) 3244This macro allows the target to add cpu and operating system specific 3245code to the call-frame unwinder for use when there is no unwind data 3246available. The most common reason to implement this macro is to unwind 3247through signal frames. 3248 3249This macro is called from @code{uw_frame_state_for} in @file{unwind-dw2.c} 3250and @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context}; 3251@var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra} 3252for the address of the code being executed and @code{context->cfa} for 3253the stack pointer value. If the frame can be decoded, the register save 3254addresses should be updated in @var{fs} and the macro should evaluate to 3255@code{_URC_NO_REASON}. If the frame cannot be decoded, the macro should 3256evaluate to @code{_URC_END_OF_STACK}. 3257 3258For proper signal handling in Java this macro is accompanied by 3259@code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers. 3260@end defmac 3261 3262@defmac MD_HANDLE_UNWABI (@var{context}, @var{fs}) 3263This macro allows the target to add operating system specific code to the 3264call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive, 3265usually used for signal or interrupt frames. 3266 3267This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}. 3268@var{context} is an @code{_Unwind_Context}; 3269@var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi} 3270for the abi and context in the @code{.unwabi} directive. If the 3271@code{.unwabi} directive can be handled, the register save addresses should 3272be updated in @var{fs}. 3273@end defmac 3274 3275@defmac TARGET_USES_WEAK_UNWIND_INFO 3276A C expression that evaluates to true if the target requires unwind 3277info to be given comdat linkage. Define it to be @code{1} if comdat 3278linkage is necessary. The default is @code{0}. 3279@end defmac 3280 3281@node Stack Checking 3282@subsection Specifying How Stack Checking is Done 3283 3284GCC will check that stack references are within the boundaries of 3285the stack, if the @option{-fstack-check} is specified, in one of three ways: 3286 3287@enumerate 3288@item 3289If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC 3290will assume that you have arranged for stack checking to be done at 3291appropriate places in the configuration files, e.g., in 3292@code{TARGET_ASM_FUNCTION_PROLOGUE}. GCC will do not other special 3293processing. 3294 3295@item 3296If @code{STACK_CHECK_BUILTIN} is zero and you defined a named pattern 3297called @code{check_stack} in your @file{md} file, GCC will call that 3298pattern with one argument which is the address to compare the stack 3299value against. You must arrange for this pattern to report an error if 3300the stack pointer is out of range. 3301 3302@item 3303If neither of the above are true, GCC will generate code to periodically 3304``probe'' the stack pointer using the values of the macros defined below. 3305@end enumerate 3306 3307Normally, you will use the default values of these macros, so GCC 3308will use the third approach. 3309 3310@defmac STACK_CHECK_BUILTIN 3311A nonzero value if stack checking is done by the configuration files in a 3312machine-dependent manner. You should define this macro if stack checking 3313is require by the ABI of your machine or if you would like to have to stack 3314checking in some more efficient way than GCC's portable approach. 3315The default value of this macro is zero. 3316@end defmac 3317 3318@defmac STACK_CHECK_PROBE_INTERVAL 3319An integer representing the interval at which GCC must generate stack 3320probe instructions. You will normally define this macro to be no larger 3321than the size of the ``guard pages'' at the end of a stack area. The 3322default value of 4096 is suitable for most systems. 3323@end defmac 3324 3325@defmac STACK_CHECK_PROBE_LOAD 3326A integer which is nonzero if GCC should perform the stack probe 3327as a load instruction and zero if GCC should use a store instruction. 3328The default is zero, which is the most efficient choice on most systems. 3329@end defmac 3330 3331@defmac STACK_CHECK_PROTECT 3332The number of bytes of stack needed to recover from a stack overflow, 3333for languages where such a recovery is supported. The default value of 333475 words should be adequate for most machines. 3335@end defmac 3336 3337@defmac STACK_CHECK_MAX_FRAME_SIZE 3338The maximum size of a stack frame, in bytes. GCC will generate probe 3339instructions in non-leaf functions to ensure at least this many bytes of 3340stack are available. If a stack frame is larger than this size, stack 3341checking will not be reliable and GCC will issue a warning. The 3342default is chosen so that GCC only generates one instruction on most 3343systems. You should normally not change the default value of this macro. 3344@end defmac 3345 3346@defmac STACK_CHECK_FIXED_FRAME_SIZE 3347GCC uses this value to generate the above warning message. It 3348represents the amount of fixed frame used by a function, not including 3349space for any callee-saved registers, temporaries and user variables. 3350You need only specify an upper bound for this amount and will normally 3351use the default of four words. 3352@end defmac 3353 3354@defmac STACK_CHECK_MAX_VAR_SIZE 3355The maximum size, in bytes, of an object that GCC will place in the 3356fixed area of the stack frame when the user specifies 3357@option{-fstack-check}. 3358GCC computed the default from the values of the above macros and you will 3359normally not need to override that default. 3360@end defmac 3361 3362@need 2000 3363@node Frame Registers 3364@subsection Registers That Address the Stack Frame 3365 3366@c prevent bad page break with this line 3367This discusses registers that address the stack frame. 3368 3369@defmac STACK_POINTER_REGNUM 3370The register number of the stack pointer register, which must also be a 3371fixed register according to @code{FIXED_REGISTERS}. On most machines, 3372the hardware determines which register this is. 3373@end defmac 3374 3375@defmac FRAME_POINTER_REGNUM 3376The register number of the frame pointer register, which is used to 3377access automatic variables in the stack frame. On some machines, the 3378hardware determines which register this is. On other machines, you can 3379choose any register you wish for this purpose. 3380@end defmac 3381 3382@defmac HARD_FRAME_POINTER_REGNUM 3383On some machines the offset between the frame pointer and starting 3384offset of the automatic variables is not known until after register 3385allocation has been done (for example, because the saved registers are 3386between these two locations). On those machines, define 3387@code{FRAME_POINTER_REGNUM} the number of a special, fixed register to 3388be used internally until the offset is known, and define 3389@code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number 3390used for the frame pointer. 3391 3392You should define this macro only in the very rare circumstances when it 3393is not possible to calculate the offset between the frame pointer and 3394the automatic variables until after register allocation has been 3395completed. When this macro is defined, you must also indicate in your 3396definition of @code{ELIMINABLE_REGS} how to eliminate 3397@code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM} 3398or @code{STACK_POINTER_REGNUM}. 3399 3400Do not define this macro if it would be the same as 3401@code{FRAME_POINTER_REGNUM}. 3402@end defmac 3403 3404@defmac ARG_POINTER_REGNUM 3405The register number of the arg pointer register, which is used to access 3406the function's argument list. On some machines, this is the same as the 3407frame pointer register. On some machines, the hardware determines which 3408register this is. On other machines, you can choose any register you 3409wish for this purpose. If this is not the same register as the frame 3410pointer register, then you must mark it as a fixed register according to 3411@code{FIXED_REGISTERS}, or arrange to be able to eliminate it 3412(@pxref{Elimination}). 3413@end defmac 3414 3415@defmac RETURN_ADDRESS_POINTER_REGNUM 3416The register number of the return address pointer register, which is used to 3417access the current function's return address from the stack. On some 3418machines, the return address is not at a fixed offset from the frame 3419pointer or stack pointer or argument pointer. This register can be defined 3420to point to the return address on the stack, and then be converted by 3421@code{ELIMINABLE_REGS} into either the frame pointer or stack pointer. 3422 3423Do not define this macro unless there is no other way to get the return 3424address from the stack. 3425@end defmac 3426 3427@defmac STATIC_CHAIN_REGNUM 3428@defmacx STATIC_CHAIN_INCOMING_REGNUM 3429Register numbers used for passing a function's static chain pointer. If 3430register windows are used, the register number as seen by the called 3431function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register 3432number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If 3433these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need 3434not be defined. 3435 3436The static chain register need not be a fixed register. 3437 3438If the static chain is passed in memory, these macros should not be 3439defined; instead, the next two macros should be defined. 3440@end defmac 3441 3442@defmac STATIC_CHAIN 3443@defmacx STATIC_CHAIN_INCOMING 3444If the static chain is passed in memory, these macros provide rtx giving 3445@code{mem} expressions that denote where they are stored. 3446@code{STATIC_CHAIN} and @code{STATIC_CHAIN_INCOMING} give the locations 3447as seen by the calling and called functions, respectively. Often the former 3448will be at an offset from the stack pointer and the latter at an offset from 3449the frame pointer. 3450 3451@findex stack_pointer_rtx 3452@findex frame_pointer_rtx 3453@findex arg_pointer_rtx 3454The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and 3455@code{arg_pointer_rtx} will have been initialized prior to the use of these 3456macros and should be used to refer to those items. 3457 3458If the static chain is passed in a register, the two previous macros should 3459be defined instead. 3460@end defmac 3461 3462@defmac DWARF_FRAME_REGISTERS 3463This macro specifies the maximum number of hard registers that can be 3464saved in a call frame. This is used to size data structures used in 3465DWARF2 exception handling. 3466 3467Prior to GCC 3.0, this macro was needed in order to establish a stable 3468exception handling ABI in the face of adding new hard registers for ISA 3469extensions. In GCC 3.0 and later, the EH ABI is insulated from changes 3470in the number of hard registers. Nevertheless, this macro can still be 3471used to reduce the runtime memory requirements of the exception handling 3472routines, which can be substantial if the ISA contains a lot of 3473registers that are not call-saved. 3474 3475If this macro is not defined, it defaults to 3476@code{FIRST_PSEUDO_REGISTER}. 3477@end defmac 3478 3479@defmac PRE_GCC3_DWARF_FRAME_REGISTERS 3480 3481This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided 3482for backward compatibility in pre GCC 3.0 compiled code. 3483 3484If this macro is not defined, it defaults to 3485@code{DWARF_FRAME_REGISTERS}. 3486@end defmac 3487 3488@defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno}) 3489 3490Define this macro if the target's representation for dwarf registers 3491is different than the internal representation for unwind column. 3492Given a dwarf register, this macro should return the internal unwind 3493column number to use instead. 3494 3495See the PowerPC's SPE target for an example. 3496@end defmac 3497 3498@defmac DWARF_FRAME_REGNUM (@var{regno}) 3499 3500Define this macro if the target's representation for dwarf registers 3501used in .eh_frame or .debug_frame is different from that used in other 3502debug info sections. Given a GCC hard register number, this macro 3503should return the .eh_frame register number. The default is 3504@code{DBX_REGISTER_NUMBER (@var{regno})}. 3505 3506@end defmac 3507 3508@defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh}) 3509 3510Define this macro to map register numbers held in the call frame info 3511that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that 3512should be output in .debug_frame (@code{@var{for_eh}} is zero) and 3513.eh_frame (@code{@var{for_eh}} is nonzero). The default is to 3514return @code{@var{regno}}. 3515 3516@end defmac 3517 3518@node Elimination 3519@subsection Eliminating Frame Pointer and Arg Pointer 3520 3521@c prevent bad page break with this line 3522This is about eliminating the frame pointer and arg pointer. 3523 3524@defmac FRAME_POINTER_REQUIRED 3525A C expression which is nonzero if a function must have and use a frame 3526pointer. This expression is evaluated in the reload pass. If its value is 3527nonzero the function will have a frame pointer. 3528 3529The expression can in principle examine the current function and decide 3530according to the facts, but on most machines the constant 0 or the 3531constant 1 suffices. Use 0 when the machine allows code to be generated 3532with no frame pointer, and doing so saves some time or space. Use 1 3533when there is no possible advantage to avoiding a frame pointer. 3534 3535In certain cases, the compiler does not know how to produce valid code 3536without a frame pointer. The compiler recognizes those cases and 3537automatically gives the function a frame pointer regardless of what 3538@code{FRAME_POINTER_REQUIRED} says. You don't need to worry about 3539them. 3540 3541In a function that does not require a frame pointer, the frame pointer 3542register can be allocated for ordinary usage, unless you mark it as a 3543fixed register. See @code{FIXED_REGISTERS} for more information. 3544@end defmac 3545 3546@findex get_frame_size 3547@defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var}) 3548A C statement to store in the variable @var{depth-var} the difference 3549between the frame pointer and the stack pointer values immediately after 3550the function prologue. The value would be computed from information 3551such as the result of @code{get_frame_size ()} and the tables of 3552registers @code{regs_ever_live} and @code{call_used_regs}. 3553 3554If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and 3555need not be defined. Otherwise, it must be defined even if 3556@code{FRAME_POINTER_REQUIRED} is defined to always be true; in that 3557case, you may set @var{depth-var} to anything. 3558@end defmac 3559 3560@defmac ELIMINABLE_REGS 3561If defined, this macro specifies a table of register pairs used to 3562eliminate unneeded registers that point into the stack frame. If it is not 3563defined, the only elimination attempted by the compiler is to replace 3564references to the frame pointer with references to the stack pointer. 3565 3566The definition of this macro is a list of structure initializations, each 3567of which specifies an original and replacement register. 3568 3569On some machines, the position of the argument pointer is not known until 3570the compilation is completed. In such a case, a separate hard register 3571must be used for the argument pointer. This register can be eliminated by 3572replacing it with either the frame pointer or the argument pointer, 3573depending on whether or not the frame pointer has been eliminated. 3574 3575In this case, you might specify: 3576@smallexample 3577#define ELIMINABLE_REGS \ 3578@{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \ 3579 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \ 3580 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@} 3581@end smallexample 3582 3583Note that the elimination of the argument pointer with the stack pointer is 3584specified first since that is the preferred elimination. 3585@end defmac 3586 3587@defmac CAN_ELIMINATE (@var{from-reg}, @var{to-reg}) 3588A C expression that returns nonzero if the compiler is allowed to try 3589to replace register number @var{from-reg} with register number 3590@var{to-reg}. This macro need only be defined if @code{ELIMINABLE_REGS} 3591is defined, and will usually be the constant 1, since most of the cases 3592preventing register elimination are things that the compiler already 3593knows about. 3594@end defmac 3595 3596@defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var}) 3597This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It 3598specifies the initial difference between the specified pair of 3599registers. This macro must be defined if @code{ELIMINABLE_REGS} is 3600defined. 3601@end defmac 3602 3603@node Stack Arguments 3604@subsection Passing Function Arguments on the Stack 3605@cindex arguments on stack 3606@cindex stack arguments 3607 3608The macros in this section control how arguments are passed 3609on the stack. See the following section for other macros that 3610control passing certain arguments in registers. 3611 3612@deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (tree @var{fntype}) 3613This target hook returns @code{true} if an argument declared in a 3614prototype as an integral type smaller than @code{int} should actually be 3615passed as an @code{int}. In addition to avoiding errors in certain 3616cases of mismatch, it also makes for better code on certain machines. 3617The default is to not promote prototypes. 3618@end deftypefn 3619 3620@defmac PUSH_ARGS 3621A C expression. If nonzero, push insns will be used to pass 3622outgoing arguments. 3623If the target machine does not have a push instruction, set it to zero. 3624That directs GCC to use an alternate strategy: to 3625allocate the entire argument block and then store the arguments into 3626it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too. 3627@end defmac 3628 3629@defmac PUSH_ARGS_REVERSED 3630A C expression. If nonzero, function arguments will be evaluated from 3631last to first, rather than from first to last. If this macro is not 3632defined, it defaults to @code{PUSH_ARGS} on targets where the stack 3633and args grow in opposite directions, and 0 otherwise. 3634@end defmac 3635 3636@defmac PUSH_ROUNDING (@var{npushed}) 3637A C expression that is the number of bytes actually pushed onto the 3638stack when an instruction attempts to push @var{npushed} bytes. 3639 3640On some machines, the definition 3641 3642@smallexample 3643#define PUSH_ROUNDING(BYTES) (BYTES) 3644@end smallexample 3645 3646@noindent 3647will suffice. But on other machines, instructions that appear 3648to push one byte actually push two bytes in an attempt to maintain 3649alignment. Then the definition should be 3650 3651@smallexample 3652#define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1) 3653@end smallexample 3654@end defmac 3655 3656@findex current_function_outgoing_args_size 3657@defmac ACCUMULATE_OUTGOING_ARGS 3658A C expression. If nonzero, the maximum amount of space required for outgoing arguments 3659will be computed and placed into the variable 3660@code{current_function_outgoing_args_size}. No space will be pushed 3661onto the stack for each call; instead, the function prologue should 3662increase the stack frame size by this amount. 3663 3664Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS} 3665is not proper. 3666@end defmac 3667 3668@defmac REG_PARM_STACK_SPACE (@var{fndecl}) 3669Define this macro if functions should assume that stack space has been 3670allocated for arguments even when their values are passed in 3671registers. 3672 3673The value of this macro is the size, in bytes, of the area reserved for 3674arguments passed in registers for the function represented by @var{fndecl}, 3675which can be zero if GCC is calling a library function. 3676 3677This space can be allocated by the caller, or be a part of the 3678machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says 3679which. 3680@end defmac 3681@c above is overfull. not sure what to do. --mew 5feb93 did 3682@c something, not sure if it looks good. --mew 10feb93 3683 3684@defmac OUTGOING_REG_PARM_STACK_SPACE 3685Define this if it is the responsibility of the caller to allocate the area 3686reserved for arguments passed in registers. 3687 3688If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls 3689whether the space for these arguments counts in the value of 3690@code{current_function_outgoing_args_size}. 3691@end defmac 3692 3693@defmac STACK_PARMS_IN_REG_PARM_AREA 3694Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the 3695stack parameters don't skip the area specified by it. 3696@c i changed this, makes more sens and it should have taken care of the 3697@c overfull.. not as specific, tho. --mew 5feb93 3698 3699Normally, when a parameter is not passed in registers, it is placed on the 3700stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro 3701suppresses this behavior and causes the parameter to be passed on the 3702stack in its natural location. 3703@end defmac 3704 3705@defmac RETURN_POPS_ARGS (@var{fundecl}, @var{funtype}, @var{stack-size}) 3706A C expression that should indicate the number of bytes of its own 3707arguments that a function pops on returning, or 0 if the 3708function pops no arguments and the caller must therefore pop them all 3709after the function returns. 3710 3711@var{fundecl} is a C variable whose value is a tree node that describes 3712the function in question. Normally it is a node of type 3713@code{FUNCTION_DECL} that describes the declaration of the function. 3714From this you can obtain the @code{DECL_ATTRIBUTES} of the function. 3715 3716@var{funtype} is a C variable whose value is a tree node that 3717describes the function in question. Normally it is a node of type 3718@code{FUNCTION_TYPE} that describes the data type of the function. 3719From this it is possible to obtain the data types of the value and 3720arguments (if known). 3721 3722When a call to a library function is being considered, @var{fundecl} 3723will contain an identifier node for the library function. Thus, if 3724you need to distinguish among various library functions, you can do so 3725by their names. Note that ``library function'' in this context means 3726a function used to perform arithmetic, whose name is known specially 3727in the compiler and was not mentioned in the C code being compiled. 3728 3729@var{stack-size} is the number of bytes of arguments passed on the 3730stack. If a variable number of bytes is passed, it is zero, and 3731argument popping will always be the responsibility of the calling function. 3732 3733On the VAX, all functions always pop their arguments, so the definition 3734of this macro is @var{stack-size}. On the 68000, using the standard 3735calling convention, no functions pop their arguments, so the value of 3736the macro is always 0 in this case. But an alternative calling 3737convention is available in which functions that take a fixed number of 3738arguments pop them but other functions (such as @code{printf}) pop 3739nothing (the caller pops all). When this convention is in use, 3740@var{funtype} is examined to determine whether a function takes a fixed 3741number of arguments. 3742@end defmac 3743 3744@defmac CALL_POPS_ARGS (@var{cum}) 3745A C expression that should indicate the number of bytes a call sequence 3746pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS} 3747when compiling a function call. 3748 3749@var{cum} is the variable in which all arguments to the called function 3750have been accumulated. 3751 3752On certain architectures, such as the SH5, a call trampoline is used 3753that pops certain registers off the stack, depending on the arguments 3754that have been passed to the function. Since this is a property of the 3755call site, not of the called function, @code{RETURN_POPS_ARGS} is not 3756appropriate. 3757@end defmac 3758 3759@node Register Arguments 3760@subsection Passing Arguments in Registers 3761@cindex arguments in registers 3762@cindex registers arguments 3763 3764This section describes the macros which let you control how various 3765types of arguments are passed in registers or how they are arranged in 3766the stack. 3767 3768@defmac FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named}) 3769A C expression that controls whether a function argument is passed 3770in a register, and which register. 3771 3772The arguments are @var{cum}, which summarizes all the previous 3773arguments; @var{mode}, the machine mode of the argument; @var{type}, 3774the data type of the argument as a tree node or 0 if that is not known 3775(which happens for C support library functions); and @var{named}, 3776which is 1 for an ordinary argument and 0 for nameless arguments that 3777correspond to @samp{@dots{}} in the called function's prototype. 3778@var{type} can be an incomplete type if a syntax error has previously 3779occurred. 3780 3781The value of the expression is usually either a @code{reg} RTX for the 3782hard register in which to pass the argument, or zero to pass the 3783argument on the stack. 3784 3785For machines like the VAX and 68000, where normally all arguments are 3786pushed, zero suffices as a definition. 3787 3788The value of the expression can also be a @code{parallel} RTX@. This is 3789used when an argument is passed in multiple locations. The mode of the 3790@code{parallel} should be the mode of the entire argument. The 3791@code{parallel} holds any number of @code{expr_list} pairs; each one 3792describes where part of the argument is passed. In each 3793@code{expr_list} the first operand must be a @code{reg} RTX for the hard 3794register in which to pass this part of the argument, and the mode of the 3795register RTX indicates how large this part of the argument is. The 3796second operand of the @code{expr_list} is a @code{const_int} which gives 3797the offset in bytes into the entire argument of where this part starts. 3798As a special exception the first @code{expr_list} in the @code{parallel} 3799RTX may have a first operand of zero. This indicates that the entire 3800argument is also stored on the stack. 3801 3802The last time this macro is called, it is called with @code{MODE == 3803VOIDmode}, and its result is passed to the @code{call} or @code{call_value} 3804pattern as operands 2 and 3 respectively. 3805 3806@cindex @file{stdarg.h} and register arguments 3807The usual way to make the ISO library @file{stdarg.h} work on a machine 3808where some arguments are usually passed in registers, is to cause 3809nameless arguments to be passed on the stack instead. This is done 3810by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0. 3811 3812@cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG} 3813@cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG} 3814You may use the hook @code{targetm.calls.must_pass_in_stack} 3815in the definition of this macro to determine if this argument is of a 3816type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE} 3817is not defined and @code{FUNCTION_ARG} returns nonzero for such an 3818argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is 3819defined, the argument will be computed in the stack and then loaded into 3820a register. 3821@end defmac 3822 3823@deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (enum machine_mode @var{mode}, tree @var{type}) 3824This target hook should return @code{true} if we should not pass @var{type} 3825solely in registers. The file @file{expr.h} defines a 3826definition that is usually appropriate, refer to @file{expr.h} for additional 3827documentation. 3828@end deftypefn 3829 3830@defmac FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named}) 3831Define this macro if the target machine has ``register windows'', so 3832that the register in which a function sees an arguments is not 3833necessarily the same as the one in which the caller passed the 3834argument. 3835 3836For such machines, @code{FUNCTION_ARG} computes the register in which 3837the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should 3838be defined in a similar fashion to tell the function being called 3839where the arguments will arrive. 3840 3841If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG} 3842serves both purposes. 3843@end defmac 3844 3845@deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named}) 3846This target hook returns the number of bytes at the beginning of an 3847argument that must be put in registers. The value must be zero for 3848arguments that are passed entirely in registers or that are entirely 3849pushed on the stack. 3850 3851On some machines, certain arguments must be passed partially in 3852registers and partially in memory. On these machines, typically the 3853first few words of arguments are passed in registers, and the rest 3854on the stack. If a multi-word argument (a @code{double} or a 3855structure) crosses that boundary, its first few words must be passed 3856in registers and the rest must be pushed. This macro tells the 3857compiler when this occurs, and how many bytes should go in registers. 3858 3859@code{FUNCTION_ARG} for these arguments should return the first 3860register to be used by the caller for this argument; likewise 3861@code{FUNCTION_INCOMING_ARG}, for the called function. 3862@end deftypefn 3863 3864@deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named}) 3865This target hook should return @code{true} if an argument at the 3866position indicated by @var{cum} should be passed by reference. This 3867predicate is queried after target independent reasons for being 3868passed by reference, such as @code{TREE_ADDRESSABLE (type)}. 3869 3870If the hook returns true, a copy of that argument is made in memory and a 3871pointer to the argument is passed instead of the argument itself. 3872The pointer is passed in whatever way is appropriate for passing a pointer 3873to that type. 3874@end deftypefn 3875 3876@deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named}) 3877The function argument described by the parameters to this hook is 3878known to be passed by reference. The hook should return true if the 3879function argument should be copied by the callee instead of copied 3880by the caller. 3881 3882For any argument for which the hook returns true, if it can be 3883determined that the argument is not modified, then a copy need 3884not be generated. 3885 3886The default version of this hook always returns false. 3887@end deftypefn 3888 3889@defmac CUMULATIVE_ARGS 3890A C type for declaring a variable that is used as the first argument of 3891@code{FUNCTION_ARG} and other related values. For some target machines, 3892the type @code{int} suffices and can hold the number of bytes of 3893argument so far. 3894 3895There is no need to record in @code{CUMULATIVE_ARGS} anything about the 3896arguments that have been passed on the stack. The compiler has other 3897variables to keep track of that. For target machines on which all 3898arguments are passed on the stack, there is no need to store anything in 3899@code{CUMULATIVE_ARGS}; however, the data structure must exist and 3900should not be empty, so use @code{int}. 3901@end defmac 3902 3903@defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args}) 3904A C statement (sans semicolon) for initializing the variable 3905@var{cum} for the state at the beginning of the argument list. The 3906variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype} 3907is the tree node for the data type of the function which will receive 3908the args, or 0 if the args are to a compiler support library function. 3909For direct calls that are not libcalls, @var{fndecl} contain the 3910declaration node of the function. @var{fndecl} is also set when 3911@code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function 3912being compiled. @var{n_named_args} is set to the number of named 3913arguments, including a structure return address if it is passed as a 3914parameter, when making a call. When processing incoming arguments, 3915@var{n_named_args} is set to @minus{}1. 3916 3917When processing a call to a compiler support library function, 3918@var{libname} identifies which one. It is a @code{symbol_ref} rtx which 3919contains the name of the function, as a string. @var{libname} is 0 when 3920an ordinary C function call is being processed. Thus, each time this 3921macro is called, either @var{libname} or @var{fntype} is nonzero, but 3922never both of them at once. 3923@end defmac 3924 3925@defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname}) 3926Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls, 3927it gets a @code{MODE} argument instead of @var{fntype}, that would be 3928@code{NULL}. @var{indirect} would always be zero, too. If this macro 3929is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname, 39300)} is used instead. 3931@end defmac 3932 3933@defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname}) 3934Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of 3935finding the arguments for the function being compiled. If this macro is 3936undefined, @code{INIT_CUMULATIVE_ARGS} is used instead. 3937 3938The value passed for @var{libname} is always 0, since library routines 3939with special calling conventions are never compiled with GCC@. The 3940argument @var{libname} exists for symmetry with 3941@code{INIT_CUMULATIVE_ARGS}. 3942@c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe. 3943@c --mew 5feb93 i switched the order of the sentences. --mew 10feb93 3944@end defmac 3945 3946@defmac FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named}) 3947A C statement (sans semicolon) to update the summarizer variable 3948@var{cum} to advance past an argument in the argument list. The 3949values @var{mode}, @var{type} and @var{named} describe that argument. 3950Once this is done, the variable @var{cum} is suitable for analyzing 3951the @emph{following} argument with @code{FUNCTION_ARG}, etc. 3952 3953This macro need not do anything if the argument in question was passed 3954on the stack. The compiler knows how to track the amount of stack space 3955used for arguments without any special help. 3956@end defmac 3957 3958@defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type}) 3959If defined, a C expression which determines whether, and in which direction, 3960to pad out an argument with extra space. The value should be of type 3961@code{enum direction}: either @code{upward} to pad above the argument, 3962@code{downward} to pad below, or @code{none} to inhibit padding. 3963 3964The @emph{amount} of padding is always just enough to reach the next 3965multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control 3966it. 3967 3968This macro has a default definition which is right for most systems. 3969For little-endian machines, the default is to pad upward. For 3970big-endian machines, the default is to pad downward for an argument of 3971constant size shorter than an @code{int}, and upward otherwise. 3972@end defmac 3973 3974@defmac PAD_VARARGS_DOWN 3975If defined, a C expression which determines whether the default 3976implementation of va_arg will attempt to pad down before reading the 3977next argument, if that argument is smaller than its aligned space as 3978controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such 3979arguments are padded down if @code{BYTES_BIG_ENDIAN} is true. 3980@end defmac 3981 3982@defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first}) 3983Specify padding for the last element of a block move between registers and 3984memory. @var{first} is nonzero if this is the only element. Defining this 3985macro allows better control of register function parameters on big-endian 3986machines, without using @code{PARALLEL} rtl. In particular, 3987@code{MUST_PASS_IN_STACK} need not test padding and mode of types in 3988registers, as there is no longer a "wrong" part of a register; For example, 3989a three byte aggregate may be passed in the high part of a register if so 3990required. 3991@end defmac 3992 3993@defmac FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type}) 3994If defined, a C expression that gives the alignment boundary, in bits, 3995of an argument with the specified mode and type. If it is not defined, 3996@code{PARM_BOUNDARY} is used for all arguments. 3997@end defmac 3998 3999@defmac FUNCTION_ARG_REGNO_P (@var{regno}) 4000A C expression that is nonzero if @var{regno} is the number of a hard 4001register in which function arguments are sometimes passed. This does 4002@emph{not} include implicit arguments such as the static chain and 4003the structure-value address. On many machines, no registers can be 4004used for this purpose since all function arguments are pushed on the 4005stack. 4006@end defmac 4007 4008@deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (tree @var{type}) 4009This hook should return true if parameter of type @var{type} are passed 4010as two scalar parameters. By default, GCC will attempt to pack complex 4011arguments into the target's word size. Some ABIs require complex arguments 4012to be split and treated as their individual components. For example, on 4013AIX64, complex floats should be passed in a pair of floating point 4014registers, even though a complex float would fit in one 64-bit floating 4015point register. 4016 4017The default value of this hook is @code{NULL}, which is treated as always 4018false. 4019@end deftypefn 4020 4021@deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void) 4022This hook returns a type node for @code{va_list} for the target. 4023The default version of the hook returns @code{void*}. 4024@end deftypefn 4025 4026@deftypefn {Target Hook} tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree @var{valist}, tree @var{type}, tree *@var{pre_p}, tree *@var{post_p}) 4027This hook performs target-specific gimplification of 4028@code{VA_ARG_EXPR}. The first two parameters correspond to the 4029arguments to @code{va_arg}; the latter two are as in 4030@code{gimplify.c:gimplify_expr}. 4031@end deftypefn 4032 4033@deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (enum machine_mode @var{mode}) 4034Define this to return nonzero if the port can handle pointers 4035with machine mode @var{mode}. The default version of this 4036hook returns true for both @code{ptr_mode} and @code{Pmode}. 4037@end deftypefn 4038 4039@deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (enum machine_mode @var{mode}) 4040Define this to return nonzero if the port is prepared to handle 4041insns involving scalar mode @var{mode}. For a scalar mode to be 4042considered supported, all the basic arithmetic and comparisons 4043must work. 4044 4045The default version of this hook returns true for any mode 4046required to handle the basic C types (as defined by the port). 4047Included here are the double-word arithmetic supported by the 4048code in @file{optabs.c}. 4049@end deftypefn 4050 4051@deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (enum machine_mode @var{mode}) 4052Define this to return nonzero if the port is prepared to handle 4053insns involving vector mode @var{mode}. At the very least, it 4054must have move patterns for this mode. 4055@end deftypefn 4056 4057@node Scalar Return 4058@subsection How Scalar Function Values Are Returned 4059@cindex return values in registers 4060@cindex values, returned by functions 4061@cindex scalars, returned as values 4062 4063This section discusses the macros that control returning scalars as 4064values---values that can fit in registers. 4065 4066@deftypefn {Target Hook} rtx TARGET_FUNCTION_VALUE (tree @var{ret_type}, tree @var{fn_decl_or_type}, bool @var{outgoing}) 4067 4068Define this to return an RTX representing the place where a function 4069returns or receives a value of data type @var{ret_type}, a tree node 4070node representing a data type. @var{fn_decl_or_type} is a tree node 4071representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a 4072function being called. If @var{outgoing} is false, the hook should 4073compute the register in which the caller will see the return value. 4074Otherwise, the hook should return an RTX representing the place where 4075a function returns a value. 4076 4077On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant. 4078(Actually, on most machines, scalar values are returned in the same 4079place regardless of mode.) The value of the expression is usually a 4080@code{reg} RTX for the hard register where the return value is stored. 4081The value can also be a @code{parallel} RTX, if the return value is in 4082multiple places. See @code{FUNCTION_ARG} for an explanation of the 4083@code{parallel} form. 4084 4085If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply 4086the same promotion rules specified in @code{PROMOTE_MODE} if 4087@var{valtype} is a scalar type. 4088 4089If the precise function being called is known, @var{func} is a tree 4090node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null 4091pointer. This makes it possible to use a different value-returning 4092convention for specific functions when all their calls are 4093known. 4094 4095Some target machines have ``register windows'' so that the register in 4096which a function returns its value is not the same as the one in which 4097the caller sees the value. For such machines, you should return 4098different RTX depending on @var{outgoing}. 4099 4100@code{TARGET_FUNCTION_VALUE} is not used for return values with 4101aggregate data types, because these are returned in another way. See 4102@code{TARGET_STRUCT_VALUE_RTX} and related macros, below. 4103@end deftypefn 4104 4105@defmac FUNCTION_VALUE (@var{valtype}, @var{func}) 4106This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for 4107a new target instead. 4108@end defmac 4109 4110@defmac FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func}) 4111This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for 4112a new target instead. 4113@end defmac 4114 4115@defmac LIBCALL_VALUE (@var{mode}) 4116A C expression to create an RTX representing the place where a library 4117function returns a value of mode @var{mode}. If the precise function 4118being called is known, @var{func} is a tree node 4119(@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null 4120pointer. This makes it possible to use a different value-returning 4121convention for specific functions when all their calls are 4122known. 4123 4124Note that ``library function'' in this context means a compiler 4125support routine, used to perform arithmetic, whose name is known 4126specially by the compiler and was not mentioned in the C code being 4127compiled. 4128 4129The definition of @code{LIBRARY_VALUE} need not be concerned aggregate 4130data types, because none of the library functions returns such types. 4131@end defmac 4132 4133@defmac FUNCTION_VALUE_REGNO_P (@var{regno}) 4134A C expression that is nonzero if @var{regno} is the number of a hard 4135register in which the values of called function may come back. 4136 4137A register whose use for returning values is limited to serving as the 4138second of a pair (for a value of type @code{double}, say) need not be 4139recognized by this macro. So for most machines, this definition 4140suffices: 4141 4142@smallexample 4143#define FUNCTION_VALUE_REGNO_P(N) ((N) == 0) 4144@end smallexample 4145 4146If the machine has register windows, so that the caller and the called 4147function use different registers for the return value, this macro 4148should recognize only the caller's register numbers. 4149@end defmac 4150 4151@defmac APPLY_RESULT_SIZE 4152Define this macro if @samp{untyped_call} and @samp{untyped_return} 4153need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for 4154saving and restoring an arbitrary return value. 4155@end defmac 4156 4157@deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (tree @var{type}) 4158This hook should return true if values of type @var{type} are returned 4159at the most significant end of a register (in other words, if they are 4160padded at the least significant end). You can assume that @var{type} 4161is returned in a register; the caller is required to check this. 4162 4163Note that the register provided by @code{TARGET_FUNCTION_VALUE} must 4164be able to hold the complete return value. For example, if a 1-, 2- 4165or 3-byte structure is returned at the most significant end of a 41664-byte register, @code{TARGET_FUNCTION_VALUE} should provide an 4167@code{SImode} rtx. 4168@end deftypefn 4169 4170@node Aggregate Return 4171@subsection How Large Values Are Returned 4172@cindex aggregates as return values 4173@cindex large return values 4174@cindex returning aggregate values 4175@cindex structure value address 4176 4177When a function value's mode is @code{BLKmode} (and in some other 4178cases), the value is not returned according to 4179@code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the 4180caller passes the address of a block of memory in which the value 4181should be stored. This address is called the @dfn{structure value 4182address}. 4183 4184This section describes how to control returning structure values in 4185memory. 4186 4187@deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (tree @var{type}, tree @var{fntype}) 4188This target hook should return a nonzero value to say to return the 4189function value in memory, just as large structures are always returned. 4190Here @var{type} will be the data type of the value, and @var{fntype} 4191will be the type of the function doing the returning, or @code{NULL} for 4192libcalls. 4193 4194Note that values of mode @code{BLKmode} must be explicitly handled 4195by this function. Also, the option @option{-fpcc-struct-return} 4196takes effect regardless of this macro. On most systems, it is 4197possible to leave the hook undefined; this causes a default 4198definition to be used, whose value is the constant 1 for @code{BLKmode} 4199values, and 0 otherwise. 4200 4201Do not use this hook to indicate that structures and unions should always 4202be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN} 4203to indicate this. 4204@end deftypefn 4205 4206@defmac DEFAULT_PCC_STRUCT_RETURN 4207Define this macro to be 1 if all structure and union return values must be 4208in memory. Since this results in slower code, this should be defined 4209only if needed for compatibility with other compilers or with an ABI@. 4210If you define this macro to be 0, then the conventions used for structure 4211and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY} 4212target hook. 4213 4214If not defined, this defaults to the value 1. 4215@end defmac 4216 4217@deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming}) 4218This target hook should return the location of the structure value 4219address (normally a @code{mem} or @code{reg}), or 0 if the address is 4220passed as an ``invisible'' first argument. Note that @var{fndecl} may 4221be @code{NULL}, for libcalls. You do not need to define this target 4222hook if the address is always passed as an ``invisible'' first 4223argument. 4224 4225On some architectures the place where the structure value address 4226is found by the called function is not the same place that the 4227caller put it. This can be due to register windows, or it could 4228be because the function prologue moves it to a different place. 4229@var{incoming} is @code{1} or @code{2} when the location is needed in 4230the context of the called function, and @code{0} in the context of 4231the caller. 4232 4233If @var{incoming} is nonzero and the address is to be found on the 4234stack, return a @code{mem} which refers to the frame pointer. If 4235@var{incoming} is @code{2}, the result is being used to fetch the 4236structure value address at the beginning of a function. If you need 4237to emit adjusting code, you should do it at this point. 4238@end deftypefn 4239 4240@defmac PCC_STATIC_STRUCT_RETURN 4241Define this macro if the usual system convention on the target machine 4242for returning structures and unions is for the called function to return 4243the address of a static variable containing the value. 4244 4245Do not define this if the usual system convention is for the caller to 4246pass an address to the subroutine. 4247 4248This macro has effect in @option{-fpcc-struct-return} mode, but it does 4249nothing when you use @option{-freg-struct-return} mode. 4250@end defmac 4251 4252@node Caller Saves 4253@subsection Caller-Saves Register Allocation 4254 4255If you enable it, GCC can save registers around function calls. This 4256makes it possible to use call-clobbered registers to hold variables that 4257must live across calls. 4258 4259@defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls}) 4260A C expression to determine whether it is worthwhile to consider placing 4261a pseudo-register in a call-clobbered hard register and saving and 4262restoring it around each function call. The expression should be 1 when 4263this is worth doing, and 0 otherwise. 4264 4265If you don't define this macro, a default is used which is good on most 4266machines: @code{4 * @var{calls} < @var{refs}}. 4267@end defmac 4268 4269@defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs}) 4270A C expression specifying which mode is required for saving @var{nregs} 4271of a pseudo-register in call-clobbered hard register @var{regno}. If 4272@var{regno} is unsuitable for caller save, @code{VOIDmode} should be 4273returned. For most machines this macro need not be defined since GCC 4274will select the smallest suitable mode. 4275@end defmac 4276 4277@node Function Entry 4278@subsection Function Entry and Exit 4279@cindex function entry and exit 4280@cindex prologue 4281@cindex epilogue 4282 4283This section describes the macros that output function entry 4284(@dfn{prologue}) and exit (@dfn{epilogue}) code. 4285 4286@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size}) 4287If defined, a function that outputs the assembler code for entry to a 4288function. The prologue is responsible for setting up the stack frame, 4289initializing the frame pointer register, saving registers that must be 4290saved, and allocating @var{size} additional bytes of storage for the 4291local variables. @var{size} is an integer. @var{file} is a stdio 4292stream to which the assembler code should be output. 4293 4294The label for the beginning of the function need not be output by this 4295macro. That has already been done when the macro is run. 4296 4297@findex regs_ever_live 4298To determine which registers to save, the macro can refer to the array 4299@code{regs_ever_live}: element @var{r} is nonzero if hard register 4300@var{r} is used anywhere within the function. This implies the function 4301prologue should save register @var{r}, provided it is not one of the 4302call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use 4303@code{regs_ever_live}.) 4304 4305On machines that have ``register windows'', the function entry code does 4306not save on the stack the registers that are in the windows, even if 4307they are supposed to be preserved by function calls; instead it takes 4308appropriate steps to ``push'' the register stack, if any non-call-used 4309registers are used in the function. 4310 4311@findex frame_pointer_needed 4312On machines where functions may or may not have frame-pointers, the 4313function entry code must vary accordingly; it must set up the frame 4314pointer if one is wanted, and not otherwise. To determine whether a 4315frame pointer is in wanted, the macro can refer to the variable 4316@code{frame_pointer_needed}. The variable's value will be 1 at run 4317time in a function that needs a frame pointer. @xref{Elimination}. 4318 4319The function entry code is responsible for allocating any stack space 4320required for the function. This stack space consists of the regions 4321listed below. In most cases, these regions are allocated in the 4322order listed, with the last listed region closest to the top of the 4323stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and 4324the highest address if it is not defined). You can use a different order 4325for a machine if doing so is more convenient or required for 4326compatibility reasons. Except in cases where required by standard 4327or by a debugger, there is no reason why the stack layout used by GCC 4328need agree with that used by other compilers for a machine. 4329@end deftypefn 4330 4331@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file}) 4332If defined, a function that outputs assembler code at the end of a 4333prologue. This should be used when the function prologue is being 4334emitted as RTL, and you have some extra assembler that needs to be 4335emitted. @xref{prologue instruction pattern}. 4336@end deftypefn 4337 4338@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file}) 4339If defined, a function that outputs assembler code at the start of an 4340epilogue. This should be used when the function epilogue is being 4341emitted as RTL, and you have some extra assembler that needs to be 4342emitted. @xref{epilogue instruction pattern}. 4343@end deftypefn 4344 4345@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size}) 4346If defined, a function that outputs the assembler code for exit from a 4347function. The epilogue is responsible for restoring the saved 4348registers and stack pointer to their values when the function was 4349called, and returning control to the caller. This macro takes the 4350same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the 4351registers to restore are determined from @code{regs_ever_live} and 4352@code{CALL_USED_REGISTERS} in the same way. 4353 4354On some machines, there is a single instruction that does all the work 4355of returning from the function. On these machines, give that 4356instruction the name @samp{return} and do not define the macro 4357@code{TARGET_ASM_FUNCTION_EPILOGUE} at all. 4358 4359Do not define a pattern named @samp{return} if you want the 4360@code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target 4361switches to control whether return instructions or epilogues are used, 4362define a @samp{return} pattern with a validity condition that tests the 4363target switches appropriately. If the @samp{return} pattern's validity 4364condition is false, epilogues will be used. 4365 4366On machines where functions may or may not have frame-pointers, the 4367function exit code must vary accordingly. Sometimes the code for these 4368two cases is completely different. To determine whether a frame pointer 4369is wanted, the macro can refer to the variable 4370@code{frame_pointer_needed}. The variable's value will be 1 when compiling 4371a function that needs a frame pointer. 4372 4373Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and 4374@code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially. 4375The C variable @code{current_function_is_leaf} is nonzero for such a 4376function. @xref{Leaf Functions}. 4377 4378On some machines, some functions pop their arguments on exit while 4379others leave that for the caller to do. For example, the 68020 when 4380given @option{-mrtd} pops arguments in functions that take a fixed 4381number of arguments. 4382 4383@findex current_function_pops_args 4384Your definition of the macro @code{RETURN_POPS_ARGS} decides which 4385functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE} 4386needs to know what was decided. The variable that is called 4387@code{current_function_pops_args} is the number of bytes of its 4388arguments that a function should pop. @xref{Scalar Return}. 4389@c what is the "its arguments" in the above sentence referring to, pray 4390@c tell? --mew 5feb93 4391@end deftypefn 4392 4393@itemize @bullet 4394@item 4395@findex current_function_pretend_args_size 4396A region of @code{current_function_pretend_args_size} bytes of 4397uninitialized space just underneath the first argument arriving on the 4398stack. (This may not be at the very start of the allocated stack region 4399if the calling sequence has pushed anything else since pushing the stack 4400arguments. But usually, on such machines, nothing else has been pushed 4401yet, because the function prologue itself does all the pushing.) This 4402region is used on machines where an argument may be passed partly in 4403registers and partly in memory, and, in some cases to support the 4404features in @code{<stdarg.h>}. 4405 4406@item 4407An area of memory used to save certain registers used by the function. 4408The size of this area, which may also include space for such things as 4409the return address and pointers to previous stack frames, is 4410machine-specific and usually depends on which registers have been used 4411in the function. Machines with register windows often do not require 4412a save area. 4413 4414@item 4415A region of at least @var{size} bytes, possibly rounded up to an allocation 4416boundary, to contain the local variables of the function. On some machines, 4417this region and the save area may occur in the opposite order, with the 4418save area closer to the top of the stack. 4419 4420@item 4421@cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames 4422Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of 4423@code{current_function_outgoing_args_size} bytes to be used for outgoing 4424argument lists of the function. @xref{Stack Arguments}. 4425@end itemize 4426 4427@defmac EXIT_IGNORE_STACK 4428Define this macro as a C expression that is nonzero if the return 4429instruction or the function epilogue ignores the value of the stack 4430pointer; in other words, if it is safe to delete an instruction to 4431adjust the stack pointer before a return from the function. The 4432default is 0. 4433 4434Note that this macro's value is relevant only for functions for which 4435frame pointers are maintained. It is never safe to delete a final 4436stack adjustment in a function that has no frame pointer, and the 4437compiler knows this regardless of @code{EXIT_IGNORE_STACK}. 4438@end defmac 4439 4440@defmac EPILOGUE_USES (@var{regno}) 4441Define this macro as a C expression that is nonzero for registers that are 4442used by the epilogue or the @samp{return} pattern. The stack and frame 4443pointer registers are already assumed to be used as needed. 4444@end defmac 4445 4446@defmac EH_USES (@var{regno}) 4447Define this macro as a C expression that is nonzero for registers that are 4448used by the exception handling mechanism, and so should be considered live 4449on entry to an exception edge. 4450@end defmac 4451 4452@defmac DELAY_SLOTS_FOR_EPILOGUE 4453Define this macro if the function epilogue contains delay slots to which 4454instructions from the rest of the function can be ``moved''. The 4455definition should be a C expression whose value is an integer 4456representing the number of delay slots there. 4457@end defmac 4458 4459@defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n}) 4460A C expression that returns 1 if @var{insn} can be placed in delay 4461slot number @var{n} of the epilogue. 4462 4463The argument @var{n} is an integer which identifies the delay slot now 4464being considered (since different slots may have different rules of 4465eligibility). It is never negative and is always less than the number 4466of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns). 4467If you reject a particular insn for a given delay slot, in principle, it 4468may be reconsidered for a subsequent delay slot. Also, other insns may 4469(at least in principle) be considered for the so far unfilled delay 4470slot. 4471 4472@findex current_function_epilogue_delay_list 4473@findex final_scan_insn 4474The insns accepted to fill the epilogue delay slots are put in an RTL 4475list made with @code{insn_list} objects, stored in the variable 4476@code{current_function_epilogue_delay_list}. The insn for the first 4477delay slot comes first in the list. Your definition of the macro 4478@code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by 4479outputting the insns in this list, usually by calling 4480@code{final_scan_insn}. 4481 4482You need not define this macro if you did not define 4483@code{DELAY_SLOTS_FOR_EPILOGUE}. 4484@end defmac 4485 4486@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_MI_THUNK (FILE *@var{file}, tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, tree @var{function}) 4487A function that outputs the assembler code for a thunk 4488function, used to implement C++ virtual function calls with multiple 4489inheritance. The thunk acts as a wrapper around a virtual function, 4490adjusting the implicit object parameter before handing control off to 4491the real function. 4492 4493First, emit code to add the integer @var{delta} to the location that 4494contains the incoming first argument. Assume that this argument 4495contains a pointer, and is the one used to pass the @code{this} pointer 4496in C++. This is the incoming argument @emph{before} the function prologue, 4497e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of 4498all other incoming arguments. 4499 4500Then, if @var{vcall_offset} is nonzero, an additional adjustment should be 4501made after adding @code{delta}. In particular, if @var{p} is the 4502adjusted pointer, the following adjustment should be made: 4503 4504@smallexample 4505p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)] 4506@end smallexample 4507 4508After the additions, emit code to jump to @var{function}, which is a 4509@code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does 4510not touch the return address. Hence returning from @var{FUNCTION} will 4511return to whoever called the current @samp{thunk}. 4512 4513The effect must be as if @var{function} had been called directly with 4514the adjusted first argument. This macro is responsible for emitting all 4515of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE} 4516and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked. 4517 4518The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function} 4519have already been extracted from it.) It might possibly be useful on 4520some targets, but probably not. 4521 4522If you do not define this macro, the target-independent code in the C++ 4523front end will generate a less efficient heavyweight thunk that calls 4524@var{function} instead of jumping to it. The generic approach does 4525not support varargs. 4526@end deftypefn 4527 4528@deftypefn {Target Hook} bool TARGET_ASM_CAN_OUTPUT_MI_THUNK (tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, tree @var{function}) 4529A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able 4530to output the assembler code for the thunk function specified by the 4531arguments it is passed, and false otherwise. In the latter case, the 4532generic approach will be used by the C++ front end, with the limitations 4533previously exposed. 4534@end deftypefn 4535 4536@node Profiling 4537@subsection Generating Code for Profiling 4538@cindex profiling, code generation 4539 4540These macros will help you generate code for profiling. 4541 4542@defmac FUNCTION_PROFILER (@var{file}, @var{labelno}) 4543A C statement or compound statement to output to @var{file} some 4544assembler code to call the profiling subroutine @code{mcount}. 4545 4546@findex mcount 4547The details of how @code{mcount} expects to be called are determined by 4548your operating system environment, not by GCC@. To figure them out, 4549compile a small program for profiling using the system's installed C 4550compiler and look at the assembler code that results. 4551 4552Older implementations of @code{mcount} expect the address of a counter 4553variable to be loaded into some register. The name of this variable is 4554@samp{LP} followed by the number @var{labelno}, so you would generate 4555the name using @samp{LP%d} in a @code{fprintf}. 4556@end defmac 4557 4558@defmac PROFILE_HOOK 4559A C statement or compound statement to output to @var{file} some assembly 4560code to call the profiling subroutine @code{mcount} even the target does 4561not support profiling. 4562@end defmac 4563 4564@defmac NO_PROFILE_COUNTERS 4565Define this macro to be an expression with a nonzero value if the 4566@code{mcount} subroutine on your system does not need a counter variable 4567allocated for each function. This is true for almost all modern 4568implementations. If you define this macro, you must not use the 4569@var{labelno} argument to @code{FUNCTION_PROFILER}. 4570@end defmac 4571 4572@defmac PROFILE_BEFORE_PROLOGUE 4573Define this macro if the code for function profiling should come before 4574the function prologue. Normally, the profiling code comes after. 4575@end defmac 4576 4577@node Tail Calls 4578@subsection Permitting tail calls 4579@cindex tail calls 4580 4581@deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp}) 4582True if it is ok to do sibling call optimization for the specified 4583call expression @var{exp}. @var{decl} will be the called function, 4584or @code{NULL} if this is an indirect call. 4585 4586It is not uncommon for limitations of calling conventions to prevent 4587tail calls to functions outside the current unit of translation, or 4588during PIC compilation. The hook is used to enforce these restrictions, 4589as the @code{sibcall} md pattern can not fail, or fall over to a 4590``normal'' call. The criteria for successful sibling call optimization 4591may vary greatly between different architectures. 4592@end deftypefn 4593 4594@deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap *@var{regs}) 4595Add any hard registers to @var{regs} that are live on entry to the 4596function. This hook only needs to be defined to provide registers that 4597cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved 4598registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM, 4599TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES, 4600FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM. 4601@end deftypefn 4602 4603@node Stack Smashing Protection 4604@subsection Stack smashing protection 4605@cindex stack smashing protection 4606 4607@deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void) 4608This hook returns a @code{DECL} node for the external variable to use 4609for the stack protection guard. This variable is initialized by the 4610runtime to some random value and is used to initialize the guard value 4611that is placed at the top of the local stack frame. The type of this 4612variable must be @code{ptr_type_node}. 4613 4614The default version of this hook creates a variable called 4615@samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}. 4616@end deftypefn 4617 4618@deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void) 4619This hook returns a tree expression that alerts the runtime that the 4620stack protect guard variable has been modified. This expression should 4621involve a call to a @code{noreturn} function. 4622 4623The default version of this hook invokes a function called 4624@samp{__stack_chk_fail}, taking no arguments. This function is 4625normally defined in @file{libgcc2.c}. 4626@end deftypefn 4627 4628@node Varargs 4629@section Implementing the Varargs Macros 4630@cindex varargs implementation 4631 4632GCC comes with an implementation of @code{<varargs.h>} and 4633@code{<stdarg.h>} that work without change on machines that pass arguments 4634on the stack. Other machines require their own implementations of 4635varargs, and the two machine independent header files must have 4636conditionals to include it. 4637 4638ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in 4639the calling convention for @code{va_start}. The traditional 4640implementation takes just one argument, which is the variable in which 4641to store the argument pointer. The ISO implementation of 4642@code{va_start} takes an additional second argument. The user is 4643supposed to write the last named argument of the function here. 4644 4645However, @code{va_start} should not use this argument. The way to find 4646the end of the named arguments is with the built-in functions described 4647below. 4648 4649@defmac __builtin_saveregs () 4650Use this built-in function to save the argument registers in memory so 4651that the varargs mechanism can access them. Both ISO and traditional 4652versions of @code{va_start} must use @code{__builtin_saveregs}, unless 4653you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead. 4654 4655On some machines, @code{__builtin_saveregs} is open-coded under the 4656control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On 4657other machines, it calls a routine written in assembler language, 4658found in @file{libgcc2.c}. 4659 4660Code generated for the call to @code{__builtin_saveregs} appears at the 4661beginning of the function, as opposed to where the call to 4662@code{__builtin_saveregs} is written, regardless of what the code is. 4663This is because the registers must be saved before the function starts 4664to use them for its own purposes. 4665@c i rewrote the first sentence above to fix an overfull hbox. --mew 4666@c 10feb93 4667@end defmac 4668 4669@defmac __builtin_args_info (@var{category}) 4670Use this built-in function to find the first anonymous arguments in 4671registers. 4672 4673In general, a machine may have several categories of registers used for 4674arguments, each for a particular category of data types. (For example, 4675on some machines, floating-point registers are used for floating-point 4676arguments while other arguments are passed in the general registers.) 4677To make non-varargs functions use the proper calling convention, you 4678have defined the @code{CUMULATIVE_ARGS} data type to record how many 4679registers in each category have been used so far 4680 4681@code{__builtin_args_info} accesses the same data structure of type 4682@code{CUMULATIVE_ARGS} after the ordinary argument layout is finished 4683with it, with @var{category} specifying which word to access. Thus, the 4684value indicates the first unused register in a given category. 4685 4686Normally, you would use @code{__builtin_args_info} in the implementation 4687of @code{va_start}, accessing each category just once and storing the 4688value in the @code{va_list} object. This is because @code{va_list} will 4689have to update the values, and there is no way to alter the 4690values accessed by @code{__builtin_args_info}. 4691@end defmac 4692 4693@defmac __builtin_next_arg (@var{lastarg}) 4694This is the equivalent of @code{__builtin_args_info}, for stack 4695arguments. It returns the address of the first anonymous stack 4696argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it 4697returns the address of the location above the first anonymous stack 4698argument. Use it in @code{va_start} to initialize the pointer for 4699fetching arguments from the stack. Also use it in @code{va_start} to 4700verify that the second parameter @var{lastarg} is the last named argument 4701of the current function. 4702@end defmac 4703 4704@defmac __builtin_classify_type (@var{object}) 4705Since each machine has its own conventions for which data types are 4706passed in which kind of register, your implementation of @code{va_arg} 4707has to embody these conventions. The easiest way to categorize the 4708specified data type is to use @code{__builtin_classify_type} together 4709with @code{sizeof} and @code{__alignof__}. 4710 4711@code{__builtin_classify_type} ignores the value of @var{object}, 4712considering only its data type. It returns an integer describing what 4713kind of type that is---integer, floating, pointer, structure, and so on. 4714 4715The file @file{typeclass.h} defines an enumeration that you can use to 4716interpret the values of @code{__builtin_classify_type}. 4717@end defmac 4718 4719These machine description macros help implement varargs: 4720 4721@deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void) 4722If defined, this hook produces the machine-specific code for a call to 4723@code{__builtin_saveregs}. This code will be moved to the very 4724beginning of the function, before any parameter access are made. The 4725return value of this function should be an RTX that contains the value 4726to use as the return of @code{__builtin_saveregs}. 4727@end deftypefn 4728 4729@deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARGS (CUMULATIVE_ARGS *@var{args_so_far}, enum machine_mode @var{mode}, tree @var{type}, int *@var{pretend_args_size}, int @var{second_time}) 4730This target hook offers an alternative to using 4731@code{__builtin_saveregs} and defining the hook 4732@code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous 4733register arguments into the stack so that all the arguments appear to 4734have been passed consecutively on the stack. Once this is done, you can 4735use the standard implementation of varargs that works for machines that 4736pass all their arguments on the stack. 4737 4738The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data 4739structure, containing the values that are obtained after processing the 4740named arguments. The arguments @var{mode} and @var{type} describe the 4741last named argument---its machine mode and its data type as a tree node. 4742 4743The target hook should do two things: first, push onto the stack all the 4744argument registers @emph{not} used for the named arguments, and second, 4745store the size of the data thus pushed into the @code{int}-valued 4746variable pointed to by @var{pretend_args_size}. The value that you 4747store here will serve as additional offset for setting up the stack 4748frame. 4749 4750Because you must generate code to push the anonymous arguments at 4751compile time without knowing their data types, 4752@code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that 4753have just a single category of argument register and use it uniformly 4754for all data types. 4755 4756If the argument @var{second_time} is nonzero, it means that the 4757arguments of the function are being analyzed for the second time. This 4758happens for an inline function, which is not actually compiled until the 4759end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should 4760not generate any instructions in this case. 4761@end deftypefn 4762 4763@deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (CUMULATIVE_ARGS *@var{ca}) 4764Define this hook to return @code{true} if the location where a function 4765argument is passed depends on whether or not it is a named argument. 4766 4767This hook controls how the @var{named} argument to @code{FUNCTION_ARG} 4768is set for varargs and stdarg functions. If this hook returns 4769@code{true}, the @var{named} argument is always true for named 4770arguments, and false for unnamed arguments. If it returns @code{false}, 4771but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true}, 4772then all arguments are treated as named. Otherwise, all named arguments 4773except the last are treated as named. 4774 4775You need not define this hook if it always returns zero. 4776@end deftypefn 4777 4778@deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED 4779If you need to conditionally change ABIs so that one works with 4780@code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither 4781@code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was 4782defined, then define this hook to return @code{true} if 4783@code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise. 4784Otherwise, you should not define this hook. 4785@end deftypefn 4786 4787@node Trampolines 4788@section Trampolines for Nested Functions 4789@cindex trampolines for nested functions 4790@cindex nested functions, trampolines for 4791 4792A @dfn{trampoline} is a small piece of code that is created at run time 4793when the address of a nested function is taken. It normally resides on 4794the stack, in the stack frame of the containing function. These macros 4795tell GCC how to generate code to allocate and initialize a 4796trampoline. 4797 4798The instructions in the trampoline must do two things: load a constant 4799address into the static chain register, and jump to the real address of 4800the nested function. On CISC machines such as the m68k, this requires 4801two instructions, a move immediate and a jump. Then the two addresses 4802exist in the trampoline as word-long immediate operands. On RISC 4803machines, it is often necessary to load each address into a register in 4804two parts. Then pieces of each address form separate immediate 4805operands. 4806 4807The code generated to initialize the trampoline must store the variable 4808parts---the static chain value and the function address---into the 4809immediate operands of the instructions. On a CISC machine, this is 4810simply a matter of copying each address to a memory reference at the 4811proper offset from the start of the trampoline. On a RISC machine, it 4812may be necessary to take out pieces of the address and store them 4813separately. 4814 4815@defmac TRAMPOLINE_TEMPLATE (@var{file}) 4816A C statement to output, on the stream @var{file}, assembler code for a 4817block of data that contains the constant parts of a trampoline. This 4818code should not include a label---the label is taken care of 4819automatically. 4820 4821If you do not define this macro, it means no template is needed 4822for the target. Do not define this macro on systems where the block move 4823code to copy the trampoline into place would be larger than the code 4824to generate it on the spot. 4825@end defmac 4826 4827@defmac TRAMPOLINE_SECTION 4828Return the section into which the trampoline template is to be placed 4829(@pxref{Sections}). The default value is @code{readonly_data_section}. 4830@end defmac 4831 4832@defmac TRAMPOLINE_SIZE 4833A C expression for the size in bytes of the trampoline, as an integer. 4834@end defmac 4835 4836@defmac TRAMPOLINE_ALIGNMENT 4837Alignment required for trampolines, in bits. 4838 4839If you don't define this macro, the value of @code{BIGGEST_ALIGNMENT} 4840is used for aligning trampolines. 4841@end defmac 4842 4843@defmac INITIALIZE_TRAMPOLINE (@var{addr}, @var{fnaddr}, @var{static_chain}) 4844A C statement to initialize the variable parts of a trampoline. 4845@var{addr} is an RTX for the address of the trampoline; @var{fnaddr} is 4846an RTX for the address of the nested function; @var{static_chain} is an 4847RTX for the static chain value that should be passed to the function 4848when it is called. 4849@end defmac 4850 4851@defmac TRAMPOLINE_ADJUST_ADDRESS (@var{addr}) 4852A C statement that should perform any machine-specific adjustment in 4853the address of the trampoline. Its argument contains the address that 4854was passed to @code{INITIALIZE_TRAMPOLINE}. In case the address to be 4855used for a function call should be different from the address in which 4856the template was stored, the different address should be assigned to 4857@var{addr}. If this macro is not defined, @var{addr} will be used for 4858function calls. 4859 4860@cindex @code{TARGET_ASM_FUNCTION_EPILOGUE} and trampolines 4861@cindex @code{TARGET_ASM_FUNCTION_PROLOGUE} and trampolines 4862If this macro is not defined, by default the trampoline is allocated as 4863a stack slot. This default is right for most machines. The exceptions 4864are machines where it is impossible to execute instructions in the stack 4865area. On such machines, you may have to implement a separate stack, 4866using this macro in conjunction with @code{TARGET_ASM_FUNCTION_PROLOGUE} 4867and @code{TARGET_ASM_FUNCTION_EPILOGUE}. 4868 4869@var{fp} points to a data structure, a @code{struct function}, which 4870describes the compilation status of the immediate containing function of 4871the function which the trampoline is for. The stack slot for the 4872trampoline is in the stack frame of this containing function. Other 4873allocation strategies probably must do something analogous with this 4874information. 4875@end defmac 4876 4877Implementing trampolines is difficult on many machines because they have 4878separate instruction and data caches. Writing into a stack location 4879fails to clear the memory in the instruction cache, so when the program 4880jumps to that location, it executes the old contents. 4881 4882Here are two possible solutions. One is to clear the relevant parts of 4883the instruction cache whenever a trampoline is set up. The other is to 4884make all trampolines identical, by having them jump to a standard 4885subroutine. The former technique makes trampoline execution faster; the 4886latter makes initialization faster. 4887 4888To clear the instruction cache when a trampoline is initialized, define 4889the following macro. 4890 4891@defmac CLEAR_INSN_CACHE (@var{beg}, @var{end}) 4892If defined, expands to a C expression clearing the @emph{instruction 4893cache} in the specified interval. The definition of this macro would 4894typically be a series of @code{asm} statements. Both @var{beg} and 4895@var{end} are both pointer expressions. 4896@end defmac 4897 4898The operating system may also require the stack to be made executable 4899before calling the trampoline. To implement this requirement, define 4900the following macro. 4901 4902@defmac ENABLE_EXECUTE_STACK 4903Define this macro if certain operations must be performed before executing 4904code located on the stack. The macro should expand to a series of C 4905file-scope constructs (e.g.@: functions) and provide a unique entry point 4906named @code{__enable_execute_stack}. The target is responsible for 4907emitting calls to the entry point in the code, for example from the 4908@code{INITIALIZE_TRAMPOLINE} macro. 4909@end defmac 4910 4911To use a standard subroutine, define the following macro. In addition, 4912you must make sure that the instructions in a trampoline fill an entire 4913cache line with identical instructions, or else ensure that the 4914beginning of the trampoline code is always aligned at the same point in 4915its cache line. Look in @file{m68k.h} as a guide. 4916 4917@defmac TRANSFER_FROM_TRAMPOLINE 4918Define this macro if trampolines need a special subroutine to do their 4919work. The macro should expand to a series of @code{asm} statements 4920which will be compiled with GCC@. They go in a library function named 4921@code{__transfer_from_trampoline}. 4922 4923If you need to avoid executing the ordinary prologue code of a compiled 4924C function when you jump to the subroutine, you can do so by placing a 4925special label of your own in the assembler code. Use one @code{asm} 4926statement to generate an assembler label, and another to make the label 4927global. Then trampolines can use that label to jump directly to your 4928special assembler code. 4929@end defmac 4930 4931@node Library Calls 4932@section Implicit Calls to Library Routines 4933@cindex library subroutine names 4934@cindex @file{libgcc.a} 4935 4936@c prevent bad page break with this line 4937Here is an explanation of implicit calls to library routines. 4938 4939@defmac DECLARE_LIBRARY_RENAMES 4940This macro, if defined, should expand to a piece of C code that will get 4941expanded when compiling functions for libgcc.a. It can be used to 4942provide alternate names for GCC's internal library functions if there 4943are ABI-mandated names that the compiler should provide. 4944@end defmac 4945 4946@findex init_one_libfunc 4947@findex set_optab_libfunc 4948@deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void) 4949This hook should declare additional library routines or rename 4950existing ones, using the functions @code{set_optab_libfunc} and 4951@code{init_one_libfunc} defined in @file{optabs.c}. 4952@code{init_optabs} calls this macro after initializing all the normal 4953library routines. 4954 4955The default is to do nothing. Most ports don't need to define this hook. 4956@end deftypefn 4957 4958@defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison}) 4959This macro should return @code{true} if the library routine that 4960implements the floating point comparison operator @var{comparison} in 4961mode @var{mode} will return a boolean, and @var{false} if it will 4962return a tristate. 4963 4964GCC's own floating point libraries return tristates from the 4965comparison operators, so the default returns false always. Most ports 4966don't need to define this macro. 4967@end defmac 4968 4969@defmac TARGET_LIB_INT_CMP_BIASED 4970This macro should evaluate to @code{true} if the integer comparison 4971functions (like @code{__cmpdi2}) return 0 to indicate that the first 4972operand is smaller than the second, 1 to indicate that they are equal, 4973and 2 to indicate that the first operand is greater than the second. 4974If this macro evaluates to @code{false} the comparison functions return 4975@minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines 4976in @file{libgcc.a}, you do not need to define this macro. 4977@end defmac 4978 4979@cindex US Software GOFAST, floating point emulation library 4980@cindex floating point emulation library, US Software GOFAST 4981@cindex GOFAST, floating point emulation library 4982@findex gofast_maybe_init_libfuncs 4983@defmac US_SOFTWARE_GOFAST 4984Define this macro if your system C library uses the US Software GOFAST 4985library to provide floating point emulation. 4986 4987In addition to defining this macro, your architecture must set 4988@code{TARGET_INIT_LIBFUNCS} to @code{gofast_maybe_init_libfuncs}, or 4989else call that function from its version of that hook. It is defined 4990in @file{config/gofast.h}, which must be included by your 4991architecture's @file{@var{cpu}.c} file. See @file{sparc/sparc.c} for 4992an example. 4993 4994If this macro is defined, the 4995@code{TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL} target hook must return 4996false for @code{SFmode} and @code{DFmode} comparisons. 4997@end defmac 4998 4999@cindex @code{EDOM}, implicit usage 5000@findex matherr 5001@defmac TARGET_EDOM 5002The value of @code{EDOM} on the target machine, as a C integer constant 5003expression. If you don't define this macro, GCC does not attempt to 5004deposit the value of @code{EDOM} into @code{errno} directly. Look in 5005@file{/usr/include/errno.h} to find the value of @code{EDOM} on your 5006system. 5007 5008If you do not define @code{TARGET_EDOM}, then compiled code reports 5009domain errors by calling the library function and letting it report the 5010error. If mathematical functions on your system use @code{matherr} when 5011there is an error, then you should leave @code{TARGET_EDOM} undefined so 5012that @code{matherr} is used normally. 5013@end defmac 5014 5015@cindex @code{errno}, implicit usage 5016@defmac GEN_ERRNO_RTX 5017Define this macro as a C expression to create an rtl expression that 5018refers to the global ``variable'' @code{errno}. (On certain systems, 5019@code{errno} may not actually be a variable.) If you don't define this 5020macro, a reasonable default is used. 5021@end defmac 5022 5023@cindex C99 math functions, implicit usage 5024@defmac TARGET_C99_FUNCTIONS 5025When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into 5026@code{sinf} and similarly for other functions defined by C99 standard. The 5027default is nonzero that should be proper value for most modern systems, however 5028number of existing systems lacks support for these functions in the runtime so 5029they needs this macro to be redefined to 0. 5030@end defmac 5031 5032@defmac NEXT_OBJC_RUNTIME 5033Define this macro to generate code for Objective-C message sending using 5034the calling convention of the NeXT system. This calling convention 5035involves passing the object, the selector and the method arguments all 5036at once to the method-lookup library function. 5037 5038The default calling convention passes just the object and the selector 5039to the lookup function, which returns a pointer to the method. 5040@end defmac 5041 5042@node Addressing Modes 5043@section Addressing Modes 5044@cindex addressing modes 5045 5046@c prevent bad page break with this line 5047This is about addressing modes. 5048 5049@defmac HAVE_PRE_INCREMENT 5050@defmacx HAVE_PRE_DECREMENT 5051@defmacx HAVE_POST_INCREMENT 5052@defmacx HAVE_POST_DECREMENT 5053A C expression that is nonzero if the machine supports pre-increment, 5054pre-decrement, post-increment, or post-decrement addressing respectively. 5055@end defmac 5056 5057@defmac HAVE_PRE_MODIFY_DISP 5058@defmacx HAVE_POST_MODIFY_DISP 5059A C expression that is nonzero if the machine supports pre- or 5060post-address side-effect generation involving constants other than 5061the size of the memory operand. 5062@end defmac 5063 5064@defmac HAVE_PRE_MODIFY_REG 5065@defmacx HAVE_POST_MODIFY_REG 5066A C expression that is nonzero if the machine supports pre- or 5067post-address side-effect generation involving a register displacement. 5068@end defmac 5069 5070@defmac CONSTANT_ADDRESS_P (@var{x}) 5071A C expression that is 1 if the RTX @var{x} is a constant which 5072is a valid address. On most machines, this can be defined as 5073@code{CONSTANT_P (@var{x})}, but a few machines are more restrictive 5074in which constant addresses are supported. 5075@end defmac 5076 5077@defmac CONSTANT_P (@var{x}) 5078@code{CONSTANT_P}, which is defined by target-independent code, 5079accepts integer-values expressions whose values are not explicitly 5080known, such as @code{symbol_ref}, @code{label_ref}, and @code{high} 5081expressions and @code{const} arithmetic expressions, in addition to 5082@code{const_int} and @code{const_double} expressions. 5083@end defmac 5084 5085@defmac MAX_REGS_PER_ADDRESS 5086A number, the maximum number of registers that can appear in a valid 5087memory address. Note that it is up to you to specify a value equal to 5088the maximum number that @code{GO_IF_LEGITIMATE_ADDRESS} would ever 5089accept. 5090@end defmac 5091 5092@defmac GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label}) 5093A C compound statement with a conditional @code{goto @var{label};} 5094executed if @var{x} (an RTX) is a legitimate memory address on the 5095target machine for a memory operand of mode @var{mode}. 5096 5097It usually pays to define several simpler macros to serve as 5098subroutines for this one. Otherwise it may be too complicated to 5099understand. 5100 5101This macro must exist in two variants: a strict variant and a 5102non-strict one. The strict variant is used in the reload pass. It 5103must be defined so that any pseudo-register that has not been 5104allocated a hard register is considered a memory reference. In 5105contexts where some kind of register is required, a pseudo-register 5106with no hard register must be rejected. 5107 5108The non-strict variant is used in other passes. It must be defined to 5109accept all pseudo-registers in every context where some kind of 5110register is required. 5111 5112@findex REG_OK_STRICT 5113Compiler source files that want to use the strict variant of this 5114macro define the macro @code{REG_OK_STRICT}. You should use an 5115@code{#ifdef REG_OK_STRICT} conditional to define the strict variant 5116in that case and the non-strict variant otherwise. 5117 5118Subroutines to check for acceptable registers for various purposes (one 5119for base registers, one for index registers, and so on) are typically 5120among the subroutines used to define @code{GO_IF_LEGITIMATE_ADDRESS}. 5121Then only these subroutine macros need have two variants; the higher 5122levels of macros may be the same whether strict or not. 5123 5124Normally, constant addresses which are the sum of a @code{symbol_ref} 5125and an integer are stored inside a @code{const} RTX to mark them as 5126constant. Therefore, there is no need to recognize such sums 5127specifically as legitimate addresses. Normally you would simply 5128recognize any @code{const} as legitimate. 5129 5130Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant 5131sums that are not marked with @code{const}. It assumes that a naked 5132@code{plus} indicates indexing. If so, then you @emph{must} reject such 5133naked constant sums as illegitimate addresses, so that none of them will 5134be given to @code{PRINT_OPERAND_ADDRESS}. 5135 5136@cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation 5137On some machines, whether a symbolic address is legitimate depends on 5138the section that the address refers to. On these machines, define the 5139target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information 5140into the @code{symbol_ref}, and then check for it here. When you see a 5141@code{const}, you will have to look inside it to find the 5142@code{symbol_ref} in order to determine the section. @xref{Assembler 5143Format}. 5144@end defmac 5145 5146@defmac FIND_BASE_TERM (@var{x}) 5147A C expression to determine the base term of address @var{x}. 5148This macro is used in only one place: `find_base_term' in alias.c. 5149 5150It is always safe for this macro to not be defined. It exists so 5151that alias analysis can understand machine-dependent addresses. 5152 5153The typical use of this macro is to handle addresses containing 5154a label_ref or symbol_ref within an UNSPEC@. 5155@end defmac 5156 5157@defmac LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win}) 5158A C compound statement that attempts to replace @var{x} with a valid 5159memory address for an operand of mode @var{mode}. @var{win} will be a 5160C statement label elsewhere in the code; the macro definition may use 5161 5162@smallexample 5163GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win}); 5164@end smallexample 5165 5166@noindent 5167to avoid further processing if the address has become legitimate. 5168 5169@findex break_out_memory_refs 5170@var{x} will always be the result of a call to @code{break_out_memory_refs}, 5171and @var{oldx} will be the operand that was given to that function to produce 5172@var{x}. 5173 5174The code generated by this macro should not alter the substructure of 5175@var{x}. If it transforms @var{x} into a more legitimate form, it 5176should assign @var{x} (which will always be a C variable) a new value. 5177 5178It is not necessary for this macro to come up with a legitimate 5179address. The compiler has standard ways of doing so in all cases. In 5180fact, it is safe to omit this macro. But often a 5181machine-dependent strategy can generate better code. 5182@end defmac 5183 5184@defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win}) 5185A C compound statement that attempts to replace @var{x}, which is an address 5186that needs reloading, with a valid memory address for an operand of mode 5187@var{mode}. @var{win} will be a C statement label elsewhere in the code. 5188It is not necessary to define this macro, but it might be useful for 5189performance reasons. 5190 5191For example, on the i386, it is sometimes possible to use a single 5192reload register instead of two by reloading a sum of two pseudo 5193registers into a register. On the other hand, for number of RISC 5194processors offsets are limited so that often an intermediate address 5195needs to be generated in order to address a stack slot. By defining 5196@code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses 5197generated for adjacent some stack slots can be made identical, and thus 5198be shared. 5199 5200@emph{Note}: This macro should be used with caution. It is necessary 5201to know something of how reload works in order to effectively use this, 5202and it is quite easy to produce macros that build in too much knowledge 5203of reload internals. 5204 5205@emph{Note}: This macro must be able to reload an address created by a 5206previous invocation of this macro. If it fails to handle such addresses 5207then the compiler may generate incorrect code or abort. 5208 5209@findex push_reload 5210The macro definition should use @code{push_reload} to indicate parts that 5211need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually 5212suitable to be passed unaltered to @code{push_reload}. 5213 5214The code generated by this macro must not alter the substructure of 5215@var{x}. If it transforms @var{x} into a more legitimate form, it 5216should assign @var{x} (which will always be a C variable) a new value. 5217This also applies to parts that you change indirectly by calling 5218@code{push_reload}. 5219 5220@findex strict_memory_address_p 5221The macro definition may use @code{strict_memory_address_p} to test if 5222the address has become legitimate. 5223 5224@findex copy_rtx 5225If you want to change only a part of @var{x}, one standard way of doing 5226this is to use @code{copy_rtx}. Note, however, that is unshares only a 5227single level of rtl. Thus, if the part to be changed is not at the 5228top level, you'll need to replace first the top level. 5229It is not necessary for this macro to come up with a legitimate 5230address; but often a machine-dependent strategy can generate better code. 5231@end defmac 5232 5233@defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label}) 5234A C statement or compound statement with a conditional @code{goto 5235@var{label};} executed if memory address @var{x} (an RTX) can have 5236different meanings depending on the machine mode of the memory 5237reference it is used for or if the address is valid for some modes 5238but not others. 5239 5240Autoincrement and autodecrement addresses typically have mode-dependent 5241effects because the amount of the increment or decrement is the size 5242of the operand being addressed. Some machines have other mode-dependent 5243addresses. Many RISC machines have no mode-dependent addresses. 5244 5245You may assume that @var{addr} is a valid address for the machine. 5246@end defmac 5247 5248@defmac LEGITIMATE_CONSTANT_P (@var{x}) 5249A C expression that is nonzero if @var{x} is a legitimate constant for 5250an immediate operand on the target machine. You can assume that 5251@var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact, 5252@samp{1} is a suitable definition for this macro on machines where 5253anything @code{CONSTANT_P} is valid. 5254@end defmac 5255 5256@deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x}) 5257This hook is used to undo the possibly obfuscating effects of the 5258@code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target 5259macros. Some backend implementations of these macros wrap symbol 5260references inside an @code{UNSPEC} rtx to represent PIC or similar 5261addressing modes. This target hook allows GCC's optimizers to understand 5262the semantics of these opaque @code{UNSPEC}s by converting them back 5263into their original form. 5264@end deftypefn 5265 5266@deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (rtx @var{x}) 5267This hook should return true if @var{x} is of a form that cannot (or 5268should not) be spilled to the constant pool. The default version of 5269this hook returns false. 5270 5271The primary reason to define this hook is to prevent reload from 5272deciding that a non-legitimate constant would be better reloaded 5273from the constant pool instead of spilling and reloading a register 5274holding the constant. This restriction is often true of addresses 5275of TLS symbols for various targets. 5276@end deftypefn 5277 5278@deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (enum machine_mode @var{mode}, rtx @var{x}) 5279This hook should return true if pool entries for constant @var{x} can 5280be placed in an @code{object_block} structure. @var{mode} is the mode 5281of @var{x}. 5282 5283The default version returns false for all constants. 5284@end deftypefn 5285 5286@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void) 5287This hook should return the DECL of a function @var{f} that given an 5288address @var{addr} as an argument returns a mask @var{m} that can be 5289used to extract from two vectors the relevant data that resides in 5290@var{addr} in case @var{addr} is not properly aligned. 5291 5292The autovectrizer, when vectorizing a load operation from an address 5293@var{addr} that may be unaligned, will generate two vector loads from 5294the two aligned addresses around @var{addr}. It then generates a 5295@code{REALIGN_LOAD} operation to extract the relevant data from the 5296two loaded vectors. The first two arguments to @code{REALIGN_LOAD}, 5297@var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and 5298the third argument, @var{OFF}, defines how the data will be extracted 5299from these two vectors: if @var{OFF} is 0, then the returned vector is 5300@var{v2}; otherwise, the returned vector is composed from the last 5301@var{VS}-@var{OFF} elements of @var{v1} concatenated to the first 5302@var{OFF} elements of @var{v2}. 5303 5304If this hook is defined, the autovectorizer will generate a call 5305to @var{f} (using the DECL tree that this hook returns) and will 5306use the return value of @var{f} as the argument @var{OFF} to 5307@code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f} 5308should comply with the semantics expected by @code{REALIGN_LOAD} 5309described above. 5310If this hook is not defined, then @var{addr} will be used as 5311the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low 5312log2(@var{VS})-1 bits of @var{addr} will be considered. 5313@end deftypefn 5314 5315@node Anchored Addresses 5316@section Anchored Addresses 5317@cindex anchored addresses 5318@cindex @option{-fsection-anchors} 5319 5320GCC usually addresses every static object as a separate entity. 5321For example, if we have: 5322 5323@smallexample 5324static int a, b, c; 5325int foo (void) @{ return a + b + c; @} 5326@end smallexample 5327 5328the code for @code{foo} will usually calculate three separate symbolic 5329addresses: those of @code{a}, @code{b} and @code{c}. On some targets, 5330it would be better to calculate just one symbolic address and access 5331the three variables relative to it. The equivalent pseudocode would 5332be something like: 5333 5334@smallexample 5335int foo (void) 5336@{ 5337 register int *xr = &x; 5338 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x]; 5339@} 5340@end smallexample 5341 5342(which isn't valid C). We refer to shared addresses like @code{x} as 5343``section anchors''. Their use is controlled by @option{-fsection-anchors}. 5344 5345The hooks below describe the target properties that GCC needs to know 5346in order to make effective use of section anchors. It won't use 5347section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET} 5348or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value. 5349 5350@deftypevar {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET 5351The minimum offset that should be applied to a section anchor. 5352On most targets, it should be the smallest offset that can be 5353applied to a base register while still giving a legitimate address 5354for every mode. The default value is 0. 5355@end deftypevar 5356 5357@deftypevar {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET 5358Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive) 5359offset that should be applied to section anchors. The default 5360value is 0. 5361@end deftypevar 5362 5363@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x}) 5364Write the assembly code to define section anchor @var{x}, which is a 5365@code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true. 5366The hook is called with the assembly output position set to the beginning 5367of @code{SYMBOL_REF_BLOCK (@var{x})}. 5368 5369If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses 5370it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}. 5371If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition 5372is @code{NULL}, which disables the use of section anchors altogether. 5373@end deftypefn 5374 5375@deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (rtx @var{x}) 5376Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF} 5377@var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and 5378@samp{!SYMBOL_REF_ANCHOR_P (@var{x})}. 5379 5380The default version is correct for most targets, but you might need to 5381intercept this hook to handle things like target-specific attributes 5382or target-specific sections. 5383@end deftypefn 5384 5385@node Condition Code 5386@section Condition Code Status 5387@cindex condition code status 5388 5389@c prevent bad page break with this line 5390This describes the condition code status. 5391 5392@findex cc_status 5393The file @file{conditions.h} defines a variable @code{cc_status} to 5394describe how the condition code was computed (in case the interpretation of 5395the condition code depends on the instruction that it was set by). This 5396variable contains the RTL expressions on which the condition code is 5397currently based, and several standard flags. 5398 5399Sometimes additional machine-specific flags must be defined in the machine 5400description header file. It can also add additional machine-specific 5401information by defining @code{CC_STATUS_MDEP}. 5402 5403@defmac CC_STATUS_MDEP 5404C code for a data type which is used for declaring the @code{mdep} 5405component of @code{cc_status}. It defaults to @code{int}. 5406 5407This macro is not used on machines that do not use @code{cc0}. 5408@end defmac 5409 5410@defmac CC_STATUS_MDEP_INIT 5411A C expression to initialize the @code{mdep} field to ``empty''. 5412The default definition does nothing, since most machines don't use 5413the field anyway. If you want to use the field, you should probably 5414define this macro to initialize it. 5415 5416This macro is not used on machines that do not use @code{cc0}. 5417@end defmac 5418 5419@defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn}) 5420A C compound statement to set the components of @code{cc_status} 5421appropriately for an insn @var{insn} whose body is @var{exp}. It is 5422this macro's responsibility to recognize insns that set the condition 5423code as a byproduct of other activity as well as those that explicitly 5424set @code{(cc0)}. 5425 5426This macro is not used on machines that do not use @code{cc0}. 5427 5428If there are insns that do not set the condition code but do alter 5429other machine registers, this macro must check to see whether they 5430invalidate the expressions that the condition code is recorded as 5431reflecting. For example, on the 68000, insns that store in address 5432registers do not set the condition code, which means that usually 5433@code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such 5434insns. But suppose that the previous insn set the condition code 5435based on location @samp{a4@@(102)} and the current insn stores a new 5436value in @samp{a4}. Although the condition code is not changed by 5437this, it will no longer be true that it reflects the contents of 5438@samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter 5439@code{cc_status} in this case to say that nothing is known about the 5440condition code value. 5441 5442The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal 5443with the results of peephole optimization: insns whose patterns are 5444@code{parallel} RTXs containing various @code{reg}, @code{mem} or 5445constants which are just the operands. The RTL structure of these 5446insns is not sufficient to indicate what the insns actually do. What 5447@code{NOTICE_UPDATE_CC} should do when it sees one is just to run 5448@code{CC_STATUS_INIT}. 5449 5450A possible definition of @code{NOTICE_UPDATE_CC} is to call a function 5451that looks at an attribute (@pxref{Insn Attributes}) named, for example, 5452@samp{cc}. This avoids having detailed information about patterns in 5453two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}. 5454@end defmac 5455 5456@defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y}) 5457Returns a mode from class @code{MODE_CC} to be used when comparison 5458operation code @var{op} is applied to rtx @var{x} and @var{y}. For 5459example, on the SPARC, @code{SELECT_CC_MODE} is defined as (see 5460@pxref{Jump Patterns} for a description of the reason for this 5461definition) 5462 5463@smallexample 5464#define SELECT_CC_MODE(OP,X,Y) \ 5465 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \ 5466 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \ 5467 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \ 5468 || GET_CODE (X) == NEG) \ 5469 ? CC_NOOVmode : CCmode)) 5470@end smallexample 5471 5472You should define this macro if and only if you define extra CC modes 5473in @file{@var{machine}-modes.def}. 5474@end defmac 5475 5476@defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1}) 5477On some machines not all possible comparisons are defined, but you can 5478convert an invalid comparison into a valid one. For example, the Alpha 5479does not have a @code{GT} comparison, but you can use an @code{LT} 5480comparison instead and swap the order of the operands. 5481 5482On such machines, define this macro to be a C statement to do any 5483required conversions. @var{code} is the initial comparison code 5484and @var{op0} and @var{op1} are the left and right operands of the 5485comparison, respectively. You should modify @var{code}, @var{op0}, and 5486@var{op1} as required. 5487 5488GCC will not assume that the comparison resulting from this macro is 5489valid but will see if the resulting insn matches a pattern in the 5490@file{md} file. 5491 5492You need not define this macro if it would never change the comparison 5493code or operands. 5494@end defmac 5495 5496@defmac REVERSIBLE_CC_MODE (@var{mode}) 5497A C expression whose value is one if it is always safe to reverse a 5498comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE} 5499can ever return @var{mode} for a floating-point inequality comparison, 5500then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero. 5501 5502You need not define this macro if it would always returns zero or if the 5503floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}. 5504For example, here is the definition used on the SPARC, where floating-point 5505inequality comparisons are always given @code{CCFPEmode}: 5506 5507@smallexample 5508#define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode) 5509@end smallexample 5510@end defmac 5511 5512@defmac REVERSE_CONDITION (@var{code}, @var{mode}) 5513A C expression whose value is reversed condition code of the @var{code} for 5514comparison done in CC_MODE @var{mode}. The macro is used only in case 5515@code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case 5516machine has some non-standard way how to reverse certain conditionals. For 5517instance in case all floating point conditions are non-trapping, compiler may 5518freely convert unordered compares to ordered one. Then definition may look 5519like: 5520 5521@smallexample 5522#define REVERSE_CONDITION(CODE, MODE) \ 5523 ((MODE) != CCFPmode ? reverse_condition (CODE) \ 5524 : reverse_condition_maybe_unordered (CODE)) 5525@end smallexample 5526@end defmac 5527 5528@defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2}) 5529A C expression that returns true if the conditional execution predicate 5530@var{op1}, a comparison operation, is the inverse of @var{op2} and vice 5531versa. Define this to return 0 if the target has conditional execution 5532predicates that cannot be reversed safely. There is no need to validate 5533that the arguments of op1 and op2 are the same, this is done separately. 5534If no expansion is specified, this macro is defined as follows: 5535 5536@smallexample 5537#define REVERSE_CONDEXEC_PREDICATES_P (x, y) \ 5538 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL)) 5539@end smallexample 5540@end defmac 5541 5542@deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *, unsigned int *) 5543On targets which do not use @code{(cc0)}, and which use a hard 5544register rather than a pseudo-register to hold condition codes, the 5545regular CSE passes are often not able to identify cases in which the 5546hard register is set to a common value. Use this hook to enable a 5547small pass which optimizes such cases. This hook should return true 5548to enable this pass, and it should set the integers to which its 5549arguments point to the hard register numbers used for condition codes. 5550When there is only one such register, as is true on most systems, the 5551integer pointed to by the second argument should be set to 5552@code{INVALID_REGNUM}. 5553 5554The default version of this hook returns false. 5555@end deftypefn 5556 5557@deftypefn {Target Hook} enum machine_mode TARGET_CC_MODES_COMPATIBLE (enum machine_mode, enum machine_mode) 5558On targets which use multiple condition code modes in class 5559@code{MODE_CC}, it is sometimes the case that a comparison can be 5560validly done in more than one mode. On such a system, define this 5561target hook to take two mode arguments and to return a mode in which 5562both comparisons may be validly done. If there is no such mode, 5563return @code{VOIDmode}. 5564 5565The default version of this hook checks whether the modes are the 5566same. If they are, it returns that mode. If they are different, it 5567returns @code{VOIDmode}. 5568@end deftypefn 5569 5570@node Costs 5571@section Describing Relative Costs of Operations 5572@cindex costs of instructions 5573@cindex relative costs 5574@cindex speed of instructions 5575 5576These macros let you describe the relative speed of various operations 5577on the target machine. 5578 5579@defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to}) 5580A C expression for the cost of moving data of mode @var{mode} from a 5581register in class @var{from} to one in class @var{to}. The classes are 5582expressed using the enumeration values such as @code{GENERAL_REGS}. A 5583value of 2 is the default; other values are interpreted relative to 5584that. 5585 5586It is not required that the cost always equal 2 when @var{from} is the 5587same as @var{to}; on some machines it is expensive to move between 5588registers if they are not general registers. 5589 5590If reload sees an insn consisting of a single @code{set} between two 5591hard registers, and if @code{REGISTER_MOVE_COST} applied to their 5592classes returns a value of 2, reload does not check to ensure that the 5593constraints of the insn are met. Setting a cost of other than 2 will 5594allow reload to verify that the constraints are met. You should do this 5595if the @samp{mov@var{m}} pattern's constraints do not allow such copying. 5596@end defmac 5597 5598@defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in}) 5599A C expression for the cost of moving data of mode @var{mode} between a 5600register of class @var{class} and memory; @var{in} is zero if the value 5601is to be written to memory, nonzero if it is to be read in. This cost 5602is relative to those in @code{REGISTER_MOVE_COST}. If moving between 5603registers and memory is more expensive than between two registers, you 5604should define this macro to express the relative cost. 5605 5606If you do not define this macro, GCC uses a default cost of 4 plus 5607the cost of copying via a secondary reload register, if one is 5608needed. If your machine requires a secondary reload register to copy 5609between memory and a register of @var{class} but the reload mechanism is 5610more complex than copying via an intermediate, define this macro to 5611reflect the actual cost of the move. 5612 5613GCC defines the function @code{memory_move_secondary_cost} if 5614secondary reloads are needed. It computes the costs due to copying via 5615a secondary register. If your machine copies from memory using a 5616secondary register in the conventional way but the default base value of 56174 is not correct for your machine, define this macro to add some other 5618value to the result of that function. The arguments to that function 5619are the same as to this macro. 5620@end defmac 5621 5622@defmac BRANCH_COST 5623A C expression for the cost of a branch instruction. A value of 1 is 5624the default; other values are interpreted relative to that. 5625@end defmac 5626 5627Here are additional macros which do not specify precise relative costs, 5628but only that certain actions are more expensive than GCC would 5629ordinarily expect. 5630 5631@defmac SLOW_BYTE_ACCESS 5632Define this macro as a C expression which is nonzero if accessing less 5633than a word of memory (i.e.@: a @code{char} or a @code{short}) is no 5634faster than accessing a word of memory, i.e., if such access 5635require more than one instruction or if there is no difference in cost 5636between byte and (aligned) word loads. 5637 5638When this macro is not defined, the compiler will access a field by 5639finding the smallest containing object; when it is defined, a fullword 5640load will be used if alignment permits. Unless bytes accesses are 5641faster than word accesses, using word accesses is preferable since it 5642may eliminate subsequent memory access if subsequent accesses occur to 5643other fields in the same word of the structure, but to different bytes. 5644@end defmac 5645 5646@defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment}) 5647Define this macro to be the value 1 if memory accesses described by the 5648@var{mode} and @var{alignment} parameters have a cost many times greater 5649than aligned accesses, for example if they are emulated in a trap 5650handler. 5651 5652When this macro is nonzero, the compiler will act as if 5653@code{STRICT_ALIGNMENT} were nonzero when generating code for block 5654moves. This can cause significantly more instructions to be produced. 5655Therefore, do not set this macro nonzero if unaligned accesses only add a 5656cycle or two to the time for a memory access. 5657 5658If the value of this macro is always zero, it need not be defined. If 5659this macro is defined, it should produce a nonzero value when 5660@code{STRICT_ALIGNMENT} is nonzero. 5661@end defmac 5662 5663@defmac MOVE_RATIO 5664The threshold of number of scalar memory-to-memory move insns, @emph{below} 5665which a sequence of insns should be generated instead of a 5666string move insn or a library call. Increasing the value will always 5667make code faster, but eventually incurs high cost in increased code size. 5668 5669Note that on machines where the corresponding move insn is a 5670@code{define_expand} that emits a sequence of insns, this macro counts 5671the number of such sequences. 5672 5673If you don't define this, a reasonable default is used. 5674@end defmac 5675 5676@defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment}) 5677A C expression used to determine whether @code{move_by_pieces} will be used to 5678copy a chunk of memory, or whether some other block move mechanism 5679will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less 5680than @code{MOVE_RATIO}. 5681@end defmac 5682 5683@defmac MOVE_MAX_PIECES 5684A C expression used by @code{move_by_pieces} to determine the largest unit 5685a load or store used to copy memory is. Defaults to @code{MOVE_MAX}. 5686@end defmac 5687 5688@defmac CLEAR_RATIO 5689The threshold of number of scalar move insns, @emph{below} which a sequence 5690of insns should be generated to clear memory instead of a string clear insn 5691or a library call. Increasing the value will always make code faster, but 5692eventually incurs high cost in increased code size. 5693 5694If you don't define this, a reasonable default is used. 5695@end defmac 5696 5697@defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment}) 5698A C expression used to determine whether @code{clear_by_pieces} will be used 5699to clear a chunk of memory, or whether some other block clear mechanism 5700will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less 5701than @code{CLEAR_RATIO}. 5702@end defmac 5703 5704@defmac STORE_BY_PIECES_P (@var{size}, @var{alignment}) 5705A C expression used to determine whether @code{store_by_pieces} will be 5706used to set a chunk of memory to a constant value, or whether some other 5707mechanism will be used. Used by @code{__builtin_memset} when storing 5708values other than constant zero and by @code{__builtin_strcpy} when 5709when called with a constant source string. 5710Defaults to 1 if @code{move_by_pieces_ninsns} returns less 5711than @code{MOVE_RATIO}. 5712@end defmac 5713 5714@defmac USE_LOAD_POST_INCREMENT (@var{mode}) 5715A C expression used to determine whether a load postincrement is a good 5716thing to use for a given mode. Defaults to the value of 5717@code{HAVE_POST_INCREMENT}. 5718@end defmac 5719 5720@defmac USE_LOAD_POST_DECREMENT (@var{mode}) 5721A C expression used to determine whether a load postdecrement is a good 5722thing to use for a given mode. Defaults to the value of 5723@code{HAVE_POST_DECREMENT}. 5724@end defmac 5725 5726@defmac USE_LOAD_PRE_INCREMENT (@var{mode}) 5727A C expression used to determine whether a load preincrement is a good 5728thing to use for a given mode. Defaults to the value of 5729@code{HAVE_PRE_INCREMENT}. 5730@end defmac 5731 5732@defmac USE_LOAD_PRE_DECREMENT (@var{mode}) 5733A C expression used to determine whether a load predecrement is a good 5734thing to use for a given mode. Defaults to the value of 5735@code{HAVE_PRE_DECREMENT}. 5736@end defmac 5737 5738@defmac USE_STORE_POST_INCREMENT (@var{mode}) 5739A C expression used to determine whether a store postincrement is a good 5740thing to use for a given mode. Defaults to the value of 5741@code{HAVE_POST_INCREMENT}. 5742@end defmac 5743 5744@defmac USE_STORE_POST_DECREMENT (@var{mode}) 5745A C expression used to determine whether a store postdecrement is a good 5746thing to use for a given mode. Defaults to the value of 5747@code{HAVE_POST_DECREMENT}. 5748@end defmac 5749 5750@defmac USE_STORE_PRE_INCREMENT (@var{mode}) 5751This macro is used to determine whether a store preincrement is a good 5752thing to use for a given mode. Defaults to the value of 5753@code{HAVE_PRE_INCREMENT}. 5754@end defmac 5755 5756@defmac USE_STORE_PRE_DECREMENT (@var{mode}) 5757This macro is used to determine whether a store predecrement is a good 5758thing to use for a given mode. Defaults to the value of 5759@code{HAVE_PRE_DECREMENT}. 5760@end defmac 5761 5762@defmac NO_FUNCTION_CSE 5763Define this macro if it is as good or better to call a constant 5764function address than to call an address kept in a register. 5765@end defmac 5766 5767@defmac RANGE_TEST_NON_SHORT_CIRCUIT 5768Define this macro if a non-short-circuit operation produced by 5769@samp{fold_range_test ()} is optimal. This macro defaults to true if 5770@code{BRANCH_COST} is greater than or equal to the value 2. 5771@end defmac 5772 5773@deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, int @var{code}, int @var{outer_code}, int *@var{total}) 5774This target hook describes the relative costs of RTL expressions. 5775 5776The cost may depend on the precise form of the expression, which is 5777available for examination in @var{x}, and the rtx code of the expression 5778in which it is contained, found in @var{outer_code}. @var{code} is the 5779expression code---redundant, since it can be obtained with 5780@code{GET_CODE (@var{x})}. 5781 5782In implementing this hook, you can use the construct 5783@code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast 5784instructions. 5785 5786On entry to the hook, @code{*@var{total}} contains a default estimate 5787for the cost of the expression. The hook should modify this value as 5788necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)} 5789for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus 5790operations, and @code{COSTS_N_INSNS (1)} for all other operations. 5791 5792When optimizing for code size, i.e.@: when @code{optimize_size} is 5793nonzero, this target hook should be used to estimate the relative 5794size cost of an expression, again relative to @code{COSTS_N_INSNS}. 5795 5796The hook returns true when all subexpressions of @var{x} have been 5797processed, and false when @code{rtx_cost} should recurse. 5798@end deftypefn 5799 5800@deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address}) 5801This hook computes the cost of an addressing mode that contains 5802@var{address}. If not defined, the cost is computed from 5803the @var{address} expression and the @code{TARGET_RTX_COST} hook. 5804 5805For most CISC machines, the default cost is a good approximation of the 5806true cost of the addressing mode. However, on RISC machines, all 5807instructions normally have the same length and execution time. Hence 5808all addresses will have equal costs. 5809 5810In cases where more than one form of an address is known, the form with 5811the lowest cost will be used. If multiple forms have the same, lowest, 5812cost, the one that is the most complex will be used. 5813 5814For example, suppose an address that is equal to the sum of a register 5815and a constant is used twice in the same basic block. When this macro 5816is not defined, the address will be computed in a register and memory 5817references will be indirect through that register. On machines where 5818the cost of the addressing mode containing the sum is no higher than 5819that of a simple indirect reference, this will produce an additional 5820instruction and possibly require an additional register. Proper 5821specification of this macro eliminates this overhead for such machines. 5822 5823This hook is never called with an invalid address. 5824 5825On machines where an address involving more than one register is as 5826cheap as an address computation involving only one register, defining 5827@code{TARGET_ADDRESS_COST} to reflect this can cause two registers to 5828be live over a region of code where only one would have been if 5829@code{TARGET_ADDRESS_COST} were not defined in that manner. This effect 5830should be considered in the definition of this macro. Equivalent costs 5831should probably only be given to addresses with different numbers of 5832registers on machines with lots of registers. 5833@end deftypefn 5834 5835@node Scheduling 5836@section Adjusting the Instruction Scheduler 5837 5838The instruction scheduler may need a fair amount of machine-specific 5839adjustment in order to produce good code. GCC provides several target 5840hooks for this purpose. It is usually enough to define just a few of 5841them: try the first ones in this list first. 5842 5843@deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void) 5844This hook returns the maximum number of instructions that can ever 5845issue at the same time on the target machine. The default is one. 5846Although the insn scheduler can define itself the possibility of issue 5847an insn on the same cycle, the value can serve as an additional 5848constraint to issue insns on the same simulated processor cycle (see 5849hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}). 5850This value must be constant over the entire compilation. If you need 5851it to vary depending on what the instructions are, you must use 5852@samp{TARGET_SCHED_VARIABLE_ISSUE}. 5853@end deftypefn 5854 5855@deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx @var{insn}, int @var{more}) 5856This hook is executed by the scheduler after it has scheduled an insn 5857from the ready list. It should return the number of insns which can 5858still be issued in the current cycle. The default is 5859@samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and 5860@code{USE}, which normally are not counted against the issue rate. 5861You should define this hook if some insns take more machine resources 5862than others, so that fewer insns can follow them in the same cycle. 5863@var{file} is either a null pointer, or a stdio stream to write any 5864debug output to. @var{verbose} is the verbose level provided by 5865@option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that 5866was scheduled. 5867@end deftypefn 5868 5869@deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx @var{insn}, rtx @var{link}, rtx @var{dep_insn}, int @var{cost}) 5870This function corrects the value of @var{cost} based on the 5871relationship between @var{insn} and @var{dep_insn} through the 5872dependence @var{link}. It should return the new value. The default 5873is to make no adjustment to @var{cost}. This can be used for example 5874to specify to the scheduler using the traditional pipeline description 5875that an output- or anti-dependence does not incur the same cost as a 5876data-dependence. If the scheduler using the automaton based pipeline 5877description, the cost of anti-dependence is zero and the cost of 5878output-dependence is maximum of one and the difference of latency 5879times of the first and the second insns. If these values are not 5880acceptable, you could use the hook to modify them too. See also 5881@pxref{Processor pipeline description}. 5882@end deftypefn 5883 5884@deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx @var{insn}, int @var{priority}) 5885This hook adjusts the integer scheduling priority @var{priority} of 5886@var{insn}. It should return the new priority. Increase the priority to 5887execute @var{insn} earlier, reduce the priority to execute @var{insn} 5888later. Do not define this hook if you do not need to adjust the 5889scheduling priorities of insns. 5890@end deftypefn 5891 5892@deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock}) 5893This hook is executed by the scheduler after it has scheduled the ready 5894list, to allow the machine description to reorder it (for example to 5895combine two small instructions together on @samp{VLIW} machines). 5896@var{file} is either a null pointer, or a stdio stream to write any 5897debug output to. @var{verbose} is the verbose level provided by 5898@option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready 5899list of instructions that are ready to be scheduled. @var{n_readyp} is 5900a pointer to the number of elements in the ready list. The scheduler 5901reads the ready list in reverse order, starting with 5902@var{ready}[@var{*n_readyp}-1] and going to @var{ready}[0]. @var{clock} 5903is the timer tick of the scheduler. You may modify the ready list and 5904the number of ready insns. The return value is the number of insns that 5905can issue this cycle; normally this is just @code{issue_rate}. See also 5906@samp{TARGET_SCHED_REORDER2}. 5907@end deftypefn 5908 5909@deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_ready}, @var{clock}) 5910Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That 5911function is called whenever the scheduler starts a new cycle. This one 5912is called once per iteration over a cycle, immediately after 5913@samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and 5914return the number of insns to be scheduled in the same cycle. Defining 5915this hook can be useful if there are frequent situations where 5916scheduling one insn causes other insns to become ready in the same 5917cycle. These other insns can then be taken into account properly. 5918@end deftypefn 5919 5920@deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx @var{head}, rtx @var{tail}) 5921This hook is called after evaluation forward dependencies of insns in 5922chain given by two parameter values (@var{head} and @var{tail} 5923correspondingly) but before insns scheduling of the insn chain. For 5924example, it can be used for better insn classification if it requires 5925analysis of dependencies. This hook can use backward and forward 5926dependencies of the insn scheduler because they are already 5927calculated. 5928@end deftypefn 5929 5930@deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready}) 5931This hook is executed by the scheduler at the beginning of each block of 5932instructions that are to be scheduled. @var{file} is either a null 5933pointer, or a stdio stream to write any debug output to. @var{verbose} 5934is the verbose level provided by @option{-fsched-verbose-@var{n}}. 5935@var{max_ready} is the maximum number of insns in the current scheduling 5936region that can be live at the same time. This can be used to allocate 5937scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}. 5938@end deftypefn 5939 5940@deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose}) 5941This hook is executed by the scheduler at the end of each block of 5942instructions that are to be scheduled. It can be used to perform 5943cleanup of any actions done by the other scheduling hooks. @var{file} 5944is either a null pointer, or a stdio stream to write any debug output 5945to. @var{verbose} is the verbose level provided by 5946@option{-fsched-verbose-@var{n}}. 5947@end deftypefn 5948 5949@deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid}) 5950This hook is executed by the scheduler after function level initializations. 5951@var{file} is either a null pointer, or a stdio stream to write any debug output to. 5952@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}. 5953@var{old_max_uid} is the maximum insn uid when scheduling begins. 5954@end deftypefn 5955 5956@deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose}) 5957This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}. 5958@var{file} is either a null pointer, or a stdio stream to write any debug output to. 5959@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}. 5960@end deftypefn 5961 5962@deftypefn {Target Hook} int TARGET_SCHED_DFA_PRE_CYCLE_INSN (void) 5963The hook returns an RTL insn. The automaton state used in the 5964pipeline hazard recognizer is changed as if the insn were scheduled 5965when the new simulated processor cycle starts. Usage of the hook may 5966simplify the automaton pipeline description for some @acronym{VLIW} 5967processors. If the hook is defined, it is used only for the automaton 5968based pipeline description. The default is not to change the state 5969when the new simulated processor cycle starts. 5970@end deftypefn 5971 5972@deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void) 5973The hook can be used to initialize data used by the previous hook. 5974@end deftypefn 5975 5976@deftypefn {Target Hook} int TARGET_SCHED_DFA_POST_CYCLE_INSN (void) 5977The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used 5978to changed the state as if the insn were scheduled when the new 5979simulated processor cycle finishes. 5980@end deftypefn 5981 5982@deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void) 5983The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but 5984used to initialize data used by the previous hook. 5985@end deftypefn 5986 5987@deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void) 5988This hook controls better choosing an insn from the ready insn queue 5989for the @acronym{DFA}-based insn scheduler. Usually the scheduler 5990chooses the first insn from the queue. If the hook returns a positive 5991value, an additional scheduler code tries all permutations of 5992@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()} 5993subsequent ready insns to choose an insn whose issue will result in 5994maximal number of issued insns on the same cycle. For the 5995@acronym{VLIW} processor, the code could actually solve the problem of 5996packing simple insns into the @acronym{VLIW} insn. Of course, if the 5997rules of @acronym{VLIW} packing are described in the automaton. 5998 5999This code also could be used for superscalar @acronym{RISC} 6000processors. Let us consider a superscalar @acronym{RISC} processor 6001with 3 pipelines. Some insns can be executed in pipelines @var{A} or 6002@var{B}, some insns can be executed only in pipelines @var{B} or 6003@var{C}, and one insn can be executed in pipeline @var{B}. The 6004processor may issue the 1st insn into @var{A} and the 2nd one into 6005@var{B}. In this case, the 3rd insn will wait for freeing @var{B} 6006until the next cycle. If the scheduler issues the 3rd insn the first, 6007the processor could issue all 3 insns per cycle. 6008 6009Actually this code demonstrates advantages of the automaton based 6010pipeline hazard recognizer. We try quickly and easy many insn 6011schedules to choose the best one. 6012 6013The default is no multipass scheduling. 6014@end deftypefn 6015 6016@deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx) 6017 6018This hook controls what insns from the ready insn queue will be 6019considered for the multipass insn scheduling. If the hook returns 6020zero for insn passed as the parameter, the insn will be not chosen to 6021be issued. 6022 6023The default is that any ready insns can be chosen to be issued. 6024@end deftypefn 6025 6026@deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *, int, rtx, int, int, int *) 6027 6028This hook is called by the insn scheduler before issuing insn passed 6029as the third parameter on given cycle. If the hook returns nonzero, 6030the insn is not issued on given processors cycle. Instead of that, 6031the processor cycle is advanced. If the value passed through the last 6032parameter is zero, the insn ready queue is not sorted on the new cycle 6033start as usually. The first parameter passes file for debugging 6034output. The second one passes the scheduler verbose level of the 6035debugging output. The forth and the fifth parameter values are 6036correspondingly processor cycle on which the previous insn has been 6037issued and the current processor cycle. 6038@end deftypefn 6039 6040@deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (rtx @var{insn1}, rtx @var{insn2}, rtx @var{dep_link}, int @var{dep_cost}, int @var{distance}) 6041This hook is used to define which dependences are considered costly by 6042the target, so costly that it is not advisable to schedule the insns that 6043are involved in the dependence too close to one another. The parameters 6044to this hook are as follows: The second parameter @var{insn2} is dependent 6045upon the first parameter @var{insn1}. The dependence between @var{insn1} 6046and @var{insn2} is represented by the third parameter @var{dep_link}. The 6047fourth parameter @var{cost} is the cost of the dependence, and the fifth 6048parameter @var{distance} is the distance in cycles between the two insns. 6049The hook returns @code{true} if considering the distance between the two 6050insns the dependence between them is considered costly by the target, 6051and @code{false} otherwise. 6052 6053Defining this hook can be useful in multiple-issue out-of-order machines, 6054where (a) it's practically hopeless to predict the actual data/resource 6055delays, however: (b) there's a better chance to predict the actual grouping 6056that will be formed, and (c) correctly emulating the grouping can be very 6057important. In such targets one may want to allow issuing dependent insns 6058closer to one another---i.e., closer than the dependence distance; however, 6059not in cases of "costly dependences", which this hooks allows to define. 6060@end deftypefn 6061 6062@deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST_2 (rtx @var{insn}, int @var{dep_type}, rtx @var{dep_insn}, int @var{cost}) 6063This hook is a modified version of @samp{TARGET_SCHED_ADJUST_COST}. Instead 6064of passing dependence as a second parameter, it passes a type of that 6065dependence. This is useful to calculate cost of dependence between insns 6066not having the corresponding link. If @samp{TARGET_SCHED_ADJUST_COST_2} is 6067defined it is used instead of @samp{TARGET_SCHED_ADJUST_COST}. 6068@end deftypefn 6069 6070@deftypefn {Target Hook} void TARGET_SCHED_H_I_D_EXTENDED (void) 6071This hook is called by the insn scheduler after emitting a new instruction to 6072the instruction stream. The hook notifies a target backend to extend its 6073per instruction data structures. 6074@end deftypefn 6075 6076@deftypefn {Target Hook} int TARGET_SCHED_SPECULATE_INSN (rtx @var{insn}, int @var{request}, rtx *@var{new_pat}) 6077This hook is called by the insn scheduler when @var{insn} has only 6078speculative dependencies and therefore can be scheduled speculatively. 6079The hook is used to check if the pattern of @var{insn} has a speculative 6080version and, in case of successful check, to generate that speculative 6081pattern. The hook should return 1, if the instruction has a speculative form, 6082or -1, if it doesn't. @var{request} describes the type of requested 6083speculation. If the return value equals 1 then @var{new_pat} is assigned 6084the generated speculative pattern. 6085@end deftypefn 6086 6087@deftypefn {Target Hook} int TARGET_SCHED_NEEDS_BLOCK_P (rtx @var{insn}) 6088This hook is called by the insn scheduler during generation of recovery code 6089for @var{insn}. It should return nonzero, if the corresponding check 6090instruction should branch to recovery code, or zero otherwise. 6091@end deftypefn 6092 6093@deftypefn {Target Hook} rtx TARGET_SCHED_GEN_CHECK (rtx @var{insn}, rtx @var{label}, int @var{mutate_p}) 6094This hook is called by the insn scheduler to generate a pattern for recovery 6095check instruction. If @var{mutate_p} is zero, then @var{insn} is a 6096speculative instruction for which the check should be generated. 6097@var{label} is either a label of a basic block, where recovery code should 6098be emitted, or a null pointer, when requested check doesn't branch to 6099recovery code (a simple check). If @var{mutate_p} is nonzero, then 6100a pattern for a branchy check corresponding to a simple check denoted by 6101@var{insn} should be generated. In this case @var{label} can't be null. 6102@end deftypefn 6103 6104@deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC (rtx @var{insn}) 6105This hook is used as a workaround for 6106@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being 6107called on the first instruction of the ready list. The hook is used to 6108discard speculative instruction that stand first in the ready list from 6109being scheduled on the current cycle. For non-speculative instructions, 6110the hook should always return nonzero. For example, in the ia64 backend 6111the hook is used to cancel data speculative insns when the ALAT table 6112is nearly full. 6113@end deftypefn 6114 6115@deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_FLAGS (unsigned int *@var{flags}, spec_info_t @var{spec_info}) 6116This hook is used by the insn scheduler to find out what features should be 6117enabled/used. @var{flags} initially may have either the SCHED_RGN or SCHED_EBB 6118bit set. This denotes the scheduler pass for which the data should be 6119provided. The target backend should modify @var{flags} by modifying 6120the bits corresponding to the following features: USE_DEPS_LIST, USE_GLAT, 6121DETACH_LIFE_INFO, and DO_SPECULATION. For the DO_SPECULATION feature 6122an additional structure @var{spec_info} should be filled by the target. 6123The structure describes speculation types that can be used in the scheduler. 6124@end deftypefn 6125 6126@node Sections 6127@section Dividing the Output into Sections (Texts, Data, @dots{}) 6128@c the above section title is WAY too long. maybe cut the part between 6129@c the (...)? --mew 10feb93 6130 6131An object file is divided into sections containing different types of 6132data. In the most common case, there are three sections: the @dfn{text 6133section}, which holds instructions and read-only data; the @dfn{data 6134section}, which holds initialized writable data; and the @dfn{bss 6135section}, which holds uninitialized data. Some systems have other kinds 6136of sections. 6137 6138@file{varasm.c} provides several well-known sections, such as 6139@code{text_section}, @code{data_section} and @code{bss_section}. 6140The normal way of controlling a @code{@var{foo}_section} variable 6141is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro, 6142as described below. The macros are only read once, when @file{varasm.c} 6143initializes itself, so their values must be run-time constants. 6144They may however depend on command-line flags. 6145 6146@emph{Note:} Some run-time files, such @file{crtstuff.c}, also make 6147use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them 6148to be string literals. 6149 6150Some assemblers require a different string to be written every time a 6151section is selected. If your assembler falls into this category, you 6152should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use 6153@code{get_unnamed_section} to set up the sections. 6154 6155You must always create a @code{text_section}, either by defining 6156@code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section} 6157in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of 6158@code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not 6159create a distinct @code{readonly_data_section}, the default is to 6160reuse @code{text_section}. 6161 6162All the other @file{varasm.c} sections are optional, and are null 6163if the target does not provide them. 6164 6165@defmac TEXT_SECTION_ASM_OP 6166A C expression whose value is a string, including spacing, containing the 6167assembler operation that should precede instructions and read-only data. 6168Normally @code{"\t.text"} is right. 6169@end defmac 6170 6171@defmac HOT_TEXT_SECTION_NAME 6172If defined, a C string constant for the name of the section containing most 6173frequently executed functions of the program. If not defined, GCC will provide 6174a default definition if the target supports named sections. 6175@end defmac 6176 6177@defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME 6178If defined, a C string constant for the name of the section containing unlikely 6179executed functions in the program. 6180@end defmac 6181 6182@defmac DATA_SECTION_ASM_OP 6183A C expression whose value is a string, including spacing, containing the 6184assembler operation to identify the following data as writable initialized 6185data. Normally @code{"\t.data"} is right. 6186@end defmac 6187 6188@defmac SDATA_SECTION_ASM_OP 6189If defined, a C expression whose value is a string, including spacing, 6190containing the assembler operation to identify the following data as 6191initialized, writable small data. 6192@end defmac 6193 6194@defmac READONLY_DATA_SECTION_ASM_OP 6195A C expression whose value is a string, including spacing, containing the 6196assembler operation to identify the following data as read-only initialized 6197data. 6198@end defmac 6199 6200@defmac BSS_SECTION_ASM_OP 6201If defined, a C expression whose value is a string, including spacing, 6202containing the assembler operation to identify the following data as 6203uninitialized global data. If not defined, and neither 6204@code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined, 6205uninitialized global data will be output in the data section if 6206@option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be 6207used. 6208@end defmac 6209 6210@defmac SBSS_SECTION_ASM_OP 6211If defined, a C expression whose value is a string, including spacing, 6212containing the assembler operation to identify the following data as 6213uninitialized, writable small data. 6214@end defmac 6215 6216@defmac INIT_SECTION_ASM_OP 6217If defined, a C expression whose value is a string, including spacing, 6218containing the assembler operation to identify the following data as 6219initialization code. If not defined, GCC will assume such a section does 6220not exist. This section has no corresponding @code{init_section} 6221variable; it is used entirely in runtime code. 6222@end defmac 6223 6224@defmac FINI_SECTION_ASM_OP 6225If defined, a C expression whose value is a string, including spacing, 6226containing the assembler operation to identify the following data as 6227finalization code. If not defined, GCC will assume such a section does 6228not exist. This section has no corresponding @code{fini_section} 6229variable; it is used entirely in runtime code. 6230@end defmac 6231 6232@defmac INIT_ARRAY_SECTION_ASM_OP 6233If defined, a C expression whose value is a string, including spacing, 6234containing the assembler operation to identify the following data as 6235part of the @code{.init_array} (or equivalent) section. If not 6236defined, GCC will assume such a section does not exist. Do not define 6237both this macro and @code{INIT_SECTION_ASM_OP}. 6238@end defmac 6239 6240@defmac FINI_ARRAY_SECTION_ASM_OP 6241If defined, a C expression whose value is a string, including spacing, 6242containing the assembler operation to identify the following data as 6243part of the @code{.fini_array} (or equivalent) section. If not 6244defined, GCC will assume such a section does not exist. Do not define 6245both this macro and @code{FINI_SECTION_ASM_OP}. 6246@end defmac 6247 6248@defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function}) 6249If defined, an ASM statement that switches to a different section 6250via @var{section_op}, calls @var{function}, and switches back to 6251the text section. This is used in @file{crtstuff.c} if 6252@code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls 6253to initialization and finalization functions from the init and fini 6254sections. By default, this macro uses a simple function call. Some 6255ports need hand-crafted assembly code to avoid dependencies on 6256registers initialized in the function prologue or to ensure that 6257constant pools don't end up too far way in the text section. 6258@end defmac 6259 6260@defmac TARGET_LIBGCC_SDATA_SECTION 6261If defined, a string which names the section into which small 6262variables defined in crtstuff and libgcc should go. This is useful 6263when the target has options for optimizing access to small data, and 6264you want the crtstuff and libgcc routines to be conservative in what 6265they expect of your application yet liberal in what your application 6266expects. For example, for targets with a @code{.sdata} section (like 6267MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't 6268require small data support from your application, but use this macro 6269to put small data into @code{.sdata} so that your application can 6270access these variables whether it uses small data or not. 6271@end defmac 6272 6273@defmac FORCE_CODE_SECTION_ALIGN 6274If defined, an ASM statement that aligns a code section to some 6275arbitrary boundary. This is used to force all fragments of the 6276@code{.init} and @code{.fini} sections to have to same alignment 6277and thus prevent the linker from having to add any padding. 6278@end defmac 6279 6280@defmac JUMP_TABLES_IN_TEXT_SECTION 6281Define this macro to be an expression with a nonzero value if jump 6282tables (for @code{tablejump} insns) should be output in the text 6283section, along with the assembler instructions. Otherwise, the 6284readonly data section is used. 6285 6286This macro is irrelevant if there is no separate readonly data section. 6287@end defmac 6288 6289@deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void) 6290Define this hook if you need to do something special to set up the 6291@file{varasm.c} sections, or if your target has some special sections 6292of its own that you need to create. 6293 6294GCC calls this hook after processing the command line, but before writing 6295any assembly code, and before calling any of the section-returning hooks 6296described below. 6297@end deftypefn 6298 6299@deftypefn {Target Hook} TARGET_ASM_RELOC_RW_MASK (void) 6300Return a mask describing how relocations should be treated when 6301selecting sections. Bit 1 should be set if global relocations 6302should be placed in a read-write section; bit 0 should be set if 6303local relocations should be placed in a read-write section. 6304 6305The default version of this function returns 3 when @option{-fpic} 6306is in effect, and 0 otherwise. The hook is typically redefined 6307when the target cannot support (some kinds of) dynamic relocations 6308in read-only sections even in executables. 6309@end deftypefn 6310 6311@deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align}) 6312Return the section into which @var{exp} should be placed. You can 6313assume that @var{exp} is either a @code{VAR_DECL} node or a constant of 6314some sort. @var{reloc} indicates whether the initial value of @var{exp} 6315requires link-time relocations. Bit 0 is set when variable contains 6316local relocations only, while bit 1 is set for global relocations. 6317@var{align} is the constant alignment in bits. 6318 6319The default version of this function takes care of putting read-only 6320variables in @code{readonly_data_section}. 6321 6322See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}. 6323@end deftypefn 6324 6325@defmac USE_SELECT_SECTION_FOR_FUNCTIONS 6326Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called 6327for @code{FUNCTION_DECL}s as well as for variables and constants. 6328 6329In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the 6330function has been determined to be likely to be called, and nonzero if 6331it is unlikely to be called. 6332@end defmac 6333 6334@deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc}) 6335Build up a unique section name, expressed as a @code{STRING_CST} node, 6336and assign it to @samp{DECL_SECTION_NAME (@var{decl})}. 6337As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether 6338the initial value of @var{exp} requires link-time relocations. 6339 6340The default version of this function appends the symbol name to the 6341ELF section name that would normally be used for the symbol. For 6342example, the function @code{foo} would be placed in @code{.text.foo}. 6343Whatever the actual target object format, this is often good enough. 6344@end deftypefn 6345 6346@deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl}) 6347Return the readonly data section associated with 6348@samp{DECL_SECTION_NAME (@var{decl})}. 6349The default version of this function selects @code{.gnu.linkonce.r.name} if 6350the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name} 6351if function is in @code{.text.name}, and the normal readonly-data section 6352otherwise. 6353@end deftypefn 6354 6355@deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (enum machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align}) 6356Return the section into which a constant @var{x}, of mode @var{mode}, 6357should be placed. You can assume that @var{x} is some kind of 6358constant in RTL@. The argument @var{mode} is redundant except in the 6359case of a @code{const_int} rtx. @var{align} is the constant alignment 6360in bits. 6361 6362The default version of this function takes care of putting symbolic 6363constants in @code{flag_pic} mode in @code{data_section} and everything 6364else in @code{readonly_data_section}. 6365@end deftypefn 6366 6367@deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p}) 6368Define this hook if references to a symbol or a constant must be 6369treated differently depending on something about the variable or 6370function named by the symbol (such as what section it is in). 6371 6372The hook is executed immediately after rtl has been created for 6373@var{decl}, which may be a variable or function declaration or 6374an entry in the constant pool. In either case, @var{rtl} is the 6375rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})} 6376in this hook; that field may not have been initialized yet. 6377 6378In the case of a constant, it is safe to assume that the rtl is 6379a @code{mem} whose address is a @code{symbol_ref}. Most decls 6380will also have this form, but that is not guaranteed. Global 6381register variables, for instance, will have a @code{reg} for their 6382rtl. (Normally the right thing to do with such unusual rtl is 6383leave it alone.) 6384 6385The @var{new_decl_p} argument will be true if this is the first time 6386that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will 6387be false for subsequent invocations, which will happen for duplicate 6388declarations. Whether or not anything must be done for the duplicate 6389declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}. 6390@var{new_decl_p} is always true when the hook is called for a constant. 6391 6392@cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO} 6393The usual thing for this hook to do is to record flags in the 6394@code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}. 6395Historically, the name string was modified if it was necessary to 6396encode more than one bit of information, but this practice is now 6397discouraged; use @code{SYMBOL_REF_FLAGS}. 6398 6399The default definition of this hook, @code{default_encode_section_info} 6400in @file{varasm.c}, sets a number of commonly-useful bits in 6401@code{SYMBOL_REF_FLAGS}. Check whether the default does what you need 6402before overriding it. 6403@end deftypefn 6404 6405@deftypefn {Target Hook} const char *TARGET_STRIP_NAME_ENCODING (const char *name) 6406Decode @var{name} and return the real name part, sans 6407the characters that @code{TARGET_ENCODE_SECTION_INFO} 6408may have added. 6409@end deftypefn 6410 6411@deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (tree @var{exp}) 6412Returns true if @var{exp} should be placed into a ``small data'' section. 6413The default version of this hook always returns false. 6414@end deftypefn 6415 6416@deftypevar {Target Hook} bool TARGET_HAVE_SRODATA_SECTION 6417Contains the value true if the target places read-only 6418``small data'' into a separate section. The default value is false. 6419@end deftypevar 6420 6421@deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (tree @var{exp}) 6422Returns true if @var{exp} names an object for which name resolution 6423rules must resolve to the current ``module'' (dynamic shared library 6424or executable image). 6425 6426The default version of this hook implements the name resolution rules 6427for ELF, which has a looser model of global name binding than other 6428currently supported object file formats. 6429@end deftypefn 6430 6431@deftypevar {Target Hook} bool TARGET_HAVE_TLS 6432Contains the value true if the target supports thread-local storage. 6433The default value is false. 6434@end deftypevar 6435 6436 6437@node PIC 6438@section Position Independent Code 6439@cindex position independent code 6440@cindex PIC 6441 6442This section describes macros that help implement generation of position 6443independent code. Simply defining these macros is not enough to 6444generate valid PIC; you must also add support to the macros 6445@code{GO_IF_LEGITIMATE_ADDRESS} and @code{PRINT_OPERAND_ADDRESS}, as 6446well as @code{LEGITIMIZE_ADDRESS}. You must modify the definition of 6447@samp{movsi} to do something appropriate when the source operand 6448contains a symbolic address. You may also need to alter the handling of 6449switch statements so that they use relative addresses. 6450@c i rearranged the order of the macros above to try to force one of 6451@c them to the next line, to eliminate an overfull hbox. --mew 10feb93 6452 6453@defmac PIC_OFFSET_TABLE_REGNUM 6454The register number of the register used to address a table of static 6455data addresses in memory. In some cases this register is defined by a 6456processor's ``application binary interface'' (ABI)@. When this macro 6457is defined, RTL is generated for this register once, as with the stack 6458pointer and frame pointer registers. If this macro is not defined, it 6459is up to the machine-dependent files to allocate such a register (if 6460necessary). Note that this register must be fixed when in use (e.g.@: 6461when @code{flag_pic} is true). 6462@end defmac 6463 6464@defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED 6465Define this macro if the register defined by 6466@code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. Do not define 6467this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined. 6468@end defmac 6469 6470@defmac LEGITIMATE_PIC_OPERAND_P (@var{x}) 6471A C expression that is nonzero if @var{x} is a legitimate immediate 6472operand on the target machine when generating position independent code. 6473You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not 6474check this. You can also assume @var{flag_pic} is true, so you need not 6475check it either. You need not define this macro if all constants 6476(including @code{SYMBOL_REF}) can be immediate operands when generating 6477position independent code. 6478@end defmac 6479 6480@node Assembler Format 6481@section Defining the Output Assembler Language 6482 6483This section describes macros whose principal purpose is to describe how 6484to write instructions in assembler language---rather than what the 6485instructions do. 6486 6487@menu 6488* File Framework:: Structural information for the assembler file. 6489* Data Output:: Output of constants (numbers, strings, addresses). 6490* Uninitialized Data:: Output of uninitialized variables. 6491* Label Output:: Output and generation of labels. 6492* Initialization:: General principles of initialization 6493 and termination routines. 6494* Macros for Initialization:: 6495 Specific macros that control the handling of 6496 initialization and termination routines. 6497* Instruction Output:: Output of actual instructions. 6498* Dispatch Tables:: Output of jump tables. 6499* Exception Region Output:: Output of exception region code. 6500* Alignment Output:: Pseudo ops for alignment and skipping data. 6501@end menu 6502 6503@node File Framework 6504@subsection The Overall Framework of an Assembler File 6505@cindex assembler format 6506@cindex output of assembler code 6507 6508@c prevent bad page break with this line 6509This describes the overall framework of an assembly file. 6510 6511@deftypefn {Target Hook} void TARGET_ASM_FILE_START () 6512@findex default_file_start 6513Output to @code{asm_out_file} any text which the assembler expects to 6514find at the beginning of a file. The default behavior is controlled 6515by two flags, documented below. Unless your target's assembler is 6516quite unusual, if you override the default, you should call 6517@code{default_file_start} at some point in your target hook. This 6518lets other target files rely on these variables. 6519@end deftypefn 6520 6521@deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF 6522If this flag is true, the text of the macro @code{ASM_APP_OFF} will be 6523printed as the very first line in the assembly file, unless 6524@option{-fverbose-asm} is in effect. (If that macro has been defined 6525to the empty string, this variable has no effect.) With the normal 6526definition of @code{ASM_APP_OFF}, the effect is to notify the GNU 6527assembler that it need not bother stripping comments or extra 6528whitespace from its input. This allows it to work a bit faster. 6529 6530The default is false. You should not set it to true unless you have 6531verified that your port does not generate any extra whitespace or 6532comments that will cause GAS to issue errors in NO_APP mode. 6533@end deftypevr 6534 6535@deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE 6536If this flag is true, @code{output_file_directive} will be called 6537for the primary source file, immediately after printing 6538@code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect 6539this to be done. The default is false. 6540@end deftypevr 6541 6542@deftypefn {Target Hook} void TARGET_ASM_FILE_END () 6543Output to @code{asm_out_file} any text which the assembler expects 6544to find at the end of a file. The default is to output nothing. 6545@end deftypefn 6546 6547@deftypefun void file_end_indicate_exec_stack () 6548Some systems use a common convention, the @samp{.note.GNU-stack} 6549special section, to indicate whether or not an object file relies on 6550the stack being executable. If your system uses this convention, you 6551should define @code{TARGET_ASM_FILE_END} to this function. If you 6552need to do other things in that hook, have your hook function call 6553this function. 6554@end deftypefun 6555 6556@defmac ASM_COMMENT_START 6557A C string constant describing how to begin a comment in the target 6558assembler language. The compiler assumes that the comment will end at 6559the end of the line. 6560@end defmac 6561 6562@defmac ASM_APP_ON 6563A C string constant for text to be output before each @code{asm} 6564statement or group of consecutive ones. Normally this is 6565@code{"#APP"}, which is a comment that has no effect on most 6566assemblers but tells the GNU assembler that it must check the lines 6567that follow for all valid assembler constructs. 6568@end defmac 6569 6570@defmac ASM_APP_OFF 6571A C string constant for text to be output after each @code{asm} 6572statement or group of consecutive ones. Normally this is 6573@code{"#NO_APP"}, which tells the GNU assembler to resume making the 6574time-saving assumptions that are valid for ordinary compiler output. 6575@end defmac 6576 6577@defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name}) 6578A C statement to output COFF information or DWARF debugging information 6579which indicates that filename @var{name} is the current source file to 6580the stdio stream @var{stream}. 6581 6582This macro need not be defined if the standard form of output 6583for the file format in use is appropriate. 6584@end defmac 6585 6586@defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string}) 6587A C statement to output the string @var{string} to the stdio stream 6588@var{stream}. If you do not call the function @code{output_quoted_string} 6589in your config files, GCC will only call it to output filenames to 6590the assembler source. So you can use it to canonicalize the format 6591of the filename using this macro. 6592@end defmac 6593 6594@defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string}) 6595A C statement to output something to the assembler file to handle a 6596@samp{#ident} directive containing the text @var{string}. If this 6597macro is not defined, nothing is output for a @samp{#ident} directive. 6598@end defmac 6599 6600@deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, unsigned int @var{align}) 6601Output assembly directives to switch to section @var{name}. The section 6602should have attributes as specified by @var{flags}, which is a bit mask 6603of the @code{SECTION_*} flags defined in @file{output.h}. If @var{align} 6604is nonzero, it contains an alignment in bytes to be used for the section, 6605otherwise some target default should be used. Only targets that must 6606specify an alignment within the section directive need pay attention to 6607@var{align} -- we will still use @code{ASM_OUTPUT_ALIGN}. 6608@end deftypefn 6609 6610@deftypefn {Target Hook} bool TARGET_HAVE_NAMED_SECTIONS 6611This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}. 6612@end deftypefn 6613 6614@anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS} 6615@deftypefn {Target Hook} bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS 6616This flag is true if we can create zeroed data by switching to a BSS 6617section and then using @code{ASM_OUTPUT_SKIP} to allocate the space. 6618This is true on most ELF targets. 6619@end deftypefn 6620 6621@deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc}) 6622Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION} 6623based on a variable or function decl, a section name, and whether or not the 6624declaration's initializer may contain runtime relocations. @var{decl} may be 6625 null, in which case read-write data should be assumed. 6626 6627The default version of this function handles choosing code vs data, 6628read-only vs read-write data, and @code{flag_pic}. You should only 6629need to override this if your target has special flags that might be 6630set via @code{__attribute__}. 6631@end deftypefn 6632 6633@need 2000 6634@node Data Output 6635@subsection Output of Data 6636 6637 6638@deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP 6639@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP 6640@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP 6641@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP 6642@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP 6643@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP 6644@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP 6645@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP 6646@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP 6647These hooks specify assembly directives for creating certain kinds 6648of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a 6649byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an 6650aligned two-byte object, and so on. Any of the hooks may be 6651@code{NULL}, indicating that no suitable directive is available. 6652 6653The compiler will print these strings at the start of a new line, 6654followed immediately by the object's initial value. In most cases, 6655the string should contain a tab, a pseudo-op, and then another tab. 6656@end deftypevr 6657 6658@deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p}) 6659The @code{assemble_integer} function uses this hook to output an 6660integer object. @var{x} is the object's value, @var{size} is its size 6661in bytes and @var{aligned_p} indicates whether it is aligned. The 6662function should return @code{true} if it was able to output the 6663object. If it returns false, @code{assemble_integer} will try to 6664split the object into smaller parts. 6665 6666The default implementation of this hook will use the 6667@code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false} 6668when the relevant string is @code{NULL}. 6669@end deftypefn 6670 6671@defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail}) 6672A C statement to recognize @var{rtx} patterns that 6673@code{output_addr_const} can't deal with, and output assembly code to 6674@var{stream} corresponding to the pattern @var{x}. This may be used to 6675allow machine-dependent @code{UNSPEC}s to appear within constants. 6676 6677If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must 6678@code{goto fail}, so that a standard error message is printed. If it 6679prints an error message itself, by calling, for example, 6680@code{output_operand_lossage}, it may just complete normally. 6681@end defmac 6682 6683@defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len}) 6684A C statement to output to the stdio stream @var{stream} an assembler 6685instruction to assemble a string constant containing the @var{len} 6686bytes at @var{ptr}. @var{ptr} will be a C expression of type 6687@code{char *} and @var{len} a C expression of type @code{int}. 6688 6689If the assembler has a @code{.ascii} pseudo-op as found in the 6690Berkeley Unix assembler, do not define the macro 6691@code{ASM_OUTPUT_ASCII}. 6692@end defmac 6693 6694@defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n}) 6695A C statement to output word @var{n} of a function descriptor for 6696@var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS} 6697is defined, and is otherwise unused. 6698@end defmac 6699 6700@defmac CONSTANT_POOL_BEFORE_FUNCTION 6701You may define this macro as a C expression. You should define the 6702expression to have a nonzero value if GCC should output the constant 6703pool for a function before the code for the function, or a zero value if 6704GCC should output the constant pool after the function. If you do 6705not define this macro, the usual case, GCC will output the constant 6706pool before the function. 6707@end defmac 6708 6709@defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size}) 6710A C statement to output assembler commands to define the start of the 6711constant pool for a function. @var{funname} is a string giving 6712the name of the function. Should the return type of the function 6713be required, it can be obtained via @var{fundecl}. @var{size} 6714is the size, in bytes, of the constant pool that will be written 6715immediately after this call. 6716 6717If no constant-pool prefix is required, the usual case, this macro need 6718not be defined. 6719@end defmac 6720 6721@defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto}) 6722A C statement (with or without semicolon) to output a constant in the 6723constant pool, if it needs special treatment. (This macro need not do 6724anything for RTL expressions that can be output normally.) 6725 6726The argument @var{file} is the standard I/O stream to output the 6727assembler code on. @var{x} is the RTL expression for the constant to 6728output, and @var{mode} is the machine mode (in case @var{x} is a 6729@samp{const_int}). @var{align} is the required alignment for the value 6730@var{x}; you should output an assembler directive to force this much 6731alignment. 6732 6733The argument @var{labelno} is a number to use in an internal label for 6734the address of this pool entry. The definition of this macro is 6735responsible for outputting the label definition at the proper place. 6736Here is how to do this: 6737 6738@smallexample 6739@code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno}); 6740@end smallexample 6741 6742When you output a pool entry specially, you should end with a 6743@code{goto} to the label @var{jumpto}. This will prevent the same pool 6744entry from being output a second time in the usual manner. 6745 6746You need not define this macro if it would do nothing. 6747@end defmac 6748 6749@defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size}) 6750A C statement to output assembler commands to at the end of the constant 6751pool for a function. @var{funname} is a string giving the name of the 6752function. Should the return type of the function be required, you can 6753obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the 6754constant pool that GCC wrote immediately before this call. 6755 6756If no constant-pool epilogue is required, the usual case, you need not 6757define this macro. 6758@end defmac 6759 6760@defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}) 6761Define this macro as a C expression which is nonzero if @var{C} is 6762used as a logical line separator by the assembler. 6763 6764If you do not define this macro, the default is that only 6765the character @samp{;} is treated as a logical line separator. 6766@end defmac 6767 6768@deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN 6769@deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN 6770These target hooks are C string constants, describing the syntax in the 6771assembler for grouping arithmetic expressions. If not overridden, they 6772default to normal parentheses, which is correct for most assemblers. 6773@end deftypevr 6774 6775 These macros are provided by @file{real.h} for writing the definitions 6776of @code{ASM_OUTPUT_DOUBLE} and the like: 6777 6778@defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l}) 6779@defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l}) 6780@defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l}) 6781@defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l}) 6782@defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l}) 6783@defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l}) 6784These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the 6785target's floating point representation, and store its bit pattern in 6786the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and 6787@code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a 6788simple @code{long int}. For the others, it should be an array of 6789@code{long int}. The number of elements in this array is determined 6790by the size of the desired target floating point data type: 32 bits of 6791it go in each @code{long int} array element. Each array element holds 679232 bits of the result, even if @code{long int} is wider than 32 bits 6793on the host machine. 6794 6795The array element values are designed so that you can print them out 6796using @code{fprintf} in the order they should appear in the target 6797machine's memory. 6798@end defmac 6799 6800@node Uninitialized Data 6801@subsection Output of Uninitialized Variables 6802 6803Each of the macros in this section is used to do the whole job of 6804outputting a single uninitialized variable. 6805 6806@defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded}) 6807A C statement (sans semicolon) to output to the stdio stream 6808@var{stream} the assembler definition of a common-label named 6809@var{name} whose size is @var{size} bytes. The variable @var{rounded} 6810is the size rounded up to whatever alignment the caller wants. 6811 6812Use the expression @code{assemble_name (@var{stream}, @var{name})} to 6813output the name itself; before and after that, output the additional 6814assembler syntax for defining the name, and a newline. 6815 6816This macro controls how the assembler definitions of uninitialized 6817common global variables are output. 6818@end defmac 6819 6820@defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment}) 6821Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a 6822separate, explicit argument. If you define this macro, it is used in 6823place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in 6824handling the required alignment of the variable. The alignment is specified 6825as the number of bits. 6826@end defmac 6827 6828@defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 6829Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the 6830variable to be output, if there is one, or @code{NULL_TREE} if there 6831is no corresponding variable. If you define this macro, GCC will use it 6832in place of both @code{ASM_OUTPUT_COMMON} and 6833@code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see 6834the variable's decl in order to chose what to output. 6835@end defmac 6836 6837@defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded}) 6838A C statement (sans semicolon) to output to the stdio stream 6839@var{stream} the assembler definition of uninitialized global @var{decl} named 6840@var{name} whose size is @var{size} bytes. The variable @var{rounded} 6841is the size rounded up to whatever alignment the caller wants. 6842 6843Try to use function @code{asm_output_bss} defined in @file{varasm.c} when 6844defining this macro. If unable, use the expression 6845@code{assemble_name (@var{stream}, @var{name})} to output the name itself; 6846before and after that, output the additional assembler syntax for defining 6847the name, and a newline. 6848 6849There are two ways of handling global BSS. One is to define either 6850this macro or its aligned counterpart, @code{ASM_OUTPUT_ALIGNED_BSS}. 6851The other is to have @code{TARGET_ASM_SELECT_SECTION} return a 6852switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}). 6853You do not need to do both. 6854 6855Some languages do not have @code{common} data, and require a 6856non-common form of global BSS in order to handle uninitialized globals 6857efficiently. C++ is one example of this. However, if the target does 6858not support global BSS, the front end may choose to make globals 6859common in order to save space in the object file. 6860@end defmac 6861 6862@defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 6863Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a 6864separate, explicit argument. If you define this macro, it is used in 6865place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in 6866handling the required alignment of the variable. The alignment is specified 6867as the number of bits. 6868 6869Try to use function @code{asm_output_aligned_bss} defined in file 6870@file{varasm.c} when defining this macro. 6871@end defmac 6872 6873@defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded}) 6874A C statement (sans semicolon) to output to the stdio stream 6875@var{stream} the assembler definition of a local-common-label named 6876@var{name} whose size is @var{size} bytes. The variable @var{rounded} 6877is the size rounded up to whatever alignment the caller wants. 6878 6879Use the expression @code{assemble_name (@var{stream}, @var{name})} to 6880output the name itself; before and after that, output the additional 6881assembler syntax for defining the name, and a newline. 6882 6883This macro controls how the assembler definitions of uninitialized 6884static variables are output. 6885@end defmac 6886 6887@defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment}) 6888Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a 6889separate, explicit argument. If you define this macro, it is used in 6890place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in 6891handling the required alignment of the variable. The alignment is specified 6892as the number of bits. 6893@end defmac 6894 6895@defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 6896Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the 6897variable to be output, if there is one, or @code{NULL_TREE} if there 6898is no corresponding variable. If you define this macro, GCC will use it 6899in place of both @code{ASM_OUTPUT_DECL} and 6900@code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see 6901the variable's decl in order to chose what to output. 6902@end defmac 6903 6904@node Label Output 6905@subsection Output and Generation of Labels 6906 6907@c prevent bad page break with this line 6908This is about outputting labels. 6909 6910@findex assemble_name 6911@defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name}) 6912A C statement (sans semicolon) to output to the stdio stream 6913@var{stream} the assembler definition of a label named @var{name}. 6914Use the expression @code{assemble_name (@var{stream}, @var{name})} to 6915output the name itself; before and after that, output the additional 6916assembler syntax for defining the name, and a newline. A default 6917definition of this macro is provided which is correct for most systems. 6918@end defmac 6919 6920@findex assemble_name_raw 6921@defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name}) 6922Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known 6923to refer to a compiler-generated label. The default definition uses 6924@code{assemble_name_raw}, which is like @code{assemble_name} except 6925that it is more efficient. 6926@end defmac 6927 6928@defmac SIZE_ASM_OP 6929A C string containing the appropriate assembler directive to specify the 6930size of a symbol, without any arguments. On systems that use ELF, the 6931default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other 6932systems, the default is not to define this macro. 6933 6934Define this macro only if it is correct to use the default definitions 6935of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE} 6936for your system. If you need your own custom definitions of those 6937macros, or if you do not need explicit symbol sizes at all, do not 6938define this macro. 6939@end defmac 6940 6941@defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size}) 6942A C statement (sans semicolon) to output to the stdio stream 6943@var{stream} a directive telling the assembler that the size of the 6944symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}. 6945If you define @code{SIZE_ASM_OP}, a default definition of this macro is 6946provided. 6947@end defmac 6948 6949@defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name}) 6950A C statement (sans semicolon) to output to the stdio stream 6951@var{stream} a directive telling the assembler to calculate the size of 6952the symbol @var{name} by subtracting its address from the current 6953address. 6954 6955If you define @code{SIZE_ASM_OP}, a default definition of this macro is 6956provided. The default assumes that the assembler recognizes a special 6957@samp{.} symbol as referring to the current address, and can calculate 6958the difference between this and another symbol. If your assembler does 6959not recognize @samp{.} or cannot do calculations with it, you will need 6960to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique. 6961@end defmac 6962 6963@defmac TYPE_ASM_OP 6964A C string containing the appropriate assembler directive to specify the 6965type of a symbol, without any arguments. On systems that use ELF, the 6966default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other 6967systems, the default is not to define this macro. 6968 6969Define this macro only if it is correct to use the default definition of 6970@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 6971custom definition of this macro, or if you do not need explicit symbol 6972types at all, do not define this macro. 6973@end defmac 6974 6975@defmac TYPE_OPERAND_FMT 6976A C string which specifies (using @code{printf} syntax) the format of 6977the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the 6978default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems, 6979the default is not to define this macro. 6980 6981Define this macro only if it is correct to use the default definition of 6982@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 6983custom definition of this macro, or if you do not need explicit symbol 6984types at all, do not define this macro. 6985@end defmac 6986 6987@defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type}) 6988A C statement (sans semicolon) to output to the stdio stream 6989@var{stream} a directive telling the assembler that the type of the 6990symbol @var{name} is @var{type}. @var{type} is a C string; currently, 6991that string is always either @samp{"function"} or @samp{"object"}, but 6992you should not count on this. 6993 6994If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default 6995definition of this macro is provided. 6996@end defmac 6997 6998@defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl}) 6999A C statement (sans semicolon) to output to the stdio stream 7000@var{stream} any text necessary for declaring the name @var{name} of a 7001function which is being defined. This macro is responsible for 7002outputting the label definition (perhaps using 7003@code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the 7004@code{FUNCTION_DECL} tree node representing the function. 7005 7006If this macro is not defined, then the function name is defined in the 7007usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 7008 7009You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition 7010of this macro. 7011@end defmac 7012 7013@defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl}) 7014A C statement (sans semicolon) to output to the stdio stream 7015@var{stream} any text necessary for declaring the size of a function 7016which is being defined. The argument @var{name} is the name of the 7017function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node 7018representing the function. 7019 7020If this macro is not defined, then the function size is not defined. 7021 7022You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition 7023of this macro. 7024@end defmac 7025 7026@defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl}) 7027A C statement (sans semicolon) to output to the stdio stream 7028@var{stream} any text necessary for declaring the name @var{name} of an 7029initialized variable which is being defined. This macro must output the 7030label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument 7031@var{decl} is the @code{VAR_DECL} tree node representing the variable. 7032 7033If this macro is not defined, then the variable name is defined in the 7034usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 7035 7036You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or 7037@code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro. 7038@end defmac 7039 7040@defmac ASM_DECLARE_CONSTANT_NAME (@var{stream}, @var{name}, @var{exp}, @var{size}) 7041A C statement (sans semicolon) to output to the stdio stream 7042@var{stream} any text necessary for declaring the name @var{name} of a 7043constant which is being defined. This macro is responsible for 7044outputting the label definition (perhaps using 7045@code{ASM_OUTPUT_LABEL}). The argument @var{exp} is the 7046value of the constant, and @var{size} is the size of the constant 7047in bytes. @var{name} will be an internal label. 7048 7049If this macro is not defined, then the @var{name} is defined in the 7050usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 7051 7052You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition 7053of this macro. 7054@end defmac 7055 7056@defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name}) 7057A C statement (sans semicolon) to output to the stdio stream 7058@var{stream} any text necessary for claiming a register @var{regno} 7059for a global variable @var{decl} with name @var{name}. 7060 7061If you don't define this macro, that is equivalent to defining it to do 7062nothing. 7063@end defmac 7064 7065@defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend}) 7066A C statement (sans semicolon) to finish up declaring a variable name 7067once the compiler has processed its initializer fully and thus has had a 7068chance to determine the size of an array when controlled by an 7069initializer. This is used on systems where it's necessary to declare 7070something about the size of the object. 7071 7072If you don't define this macro, that is equivalent to defining it to do 7073nothing. 7074 7075You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or 7076@code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro. 7077@end defmac 7078 7079@deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name}) 7080This target hook is a function to output to the stdio stream 7081@var{stream} some commands that will make the label @var{name} global; 7082that is, available for reference from other files. 7083 7084The default implementation relies on a proper definition of 7085@code{GLOBAL_ASM_OP}. 7086@end deftypefn 7087 7088@defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name}) 7089A C statement (sans semicolon) to output to the stdio stream 7090@var{stream} some commands that will make the label @var{name} weak; 7091that is, available for reference from other files but only used if 7092no other definition is available. Use the expression 7093@code{assemble_name (@var{stream}, @var{name})} to output the name 7094itself; before and after that, output the additional assembler syntax 7095for making that name weak, and a newline. 7096 7097If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not 7098support weak symbols and you should not define the @code{SUPPORTS_WEAK} 7099macro. 7100@end defmac 7101 7102@defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value}) 7103Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and 7104@code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function 7105or variable decl. If @var{value} is not @code{NULL}, this C statement 7106should output to the stdio stream @var{stream} assembler code which 7107defines (equates) the weak symbol @var{name} to have the value 7108@var{value}. If @var{value} is @code{NULL}, it should output commands 7109to make @var{name} weak. 7110@end defmac 7111 7112@defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value}) 7113Outputs a directive that enables @var{name} to be used to refer to 7114symbol @var{value} with weak-symbol semantics. @code{decl} is the 7115declaration of @code{name}. 7116@end defmac 7117 7118@defmac SUPPORTS_WEAK 7119A C expression which evaluates to true if the target supports weak symbols. 7120 7121If you don't define this macro, @file{defaults.h} provides a default 7122definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL} 7123is defined, the default definition is @samp{1}; otherwise, it is 7124@samp{0}. Define this macro if you want to control weak symbol support 7125with a compiler flag such as @option{-melf}. 7126@end defmac 7127 7128@defmac MAKE_DECL_ONE_ONLY (@var{decl}) 7129A C statement (sans semicolon) to mark @var{decl} to be emitted as a 7130public symbol such that extra copies in multiple translation units will 7131be discarded by the linker. Define this macro if your object file 7132format provides support for this concept, such as the @samp{COMDAT} 7133section flags in the Microsoft Windows PE/COFF format, and this support 7134requires changes to @var{decl}, such as putting it in a separate section. 7135@end defmac 7136 7137@defmac SUPPORTS_ONE_ONLY 7138A C expression which evaluates to true if the target supports one-only 7139semantics. 7140 7141If you don't define this macro, @file{varasm.c} provides a default 7142definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default 7143definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if 7144you want to control one-only symbol support with a compiler flag, or if 7145setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to 7146be emitted as one-only. 7147@end defmac 7148 7149@deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, const char *@var{visibility}) 7150This target hook is a function to output to @var{asm_out_file} some 7151commands that will make the symbol(s) associated with @var{decl} have 7152hidden, protected or internal visibility as specified by @var{visibility}. 7153@end deftypefn 7154 7155@defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC 7156A C expression that evaluates to true if the target's linker expects 7157that weak symbols do not appear in a static archive's table of contents. 7158The default is @code{0}. 7159 7160Leaving weak symbols out of an archive's table of contents means that, 7161if a symbol will only have a definition in one translation unit and 7162will have undefined references from other translation units, that 7163symbol should not be weak. Defining this macro to be nonzero will 7164thus have the effect that certain symbols that would normally be weak 7165(explicit template instantiations, and vtables for polymorphic classes 7166with noninline key methods) will instead be nonweak. 7167 7168The C++ ABI requires this macro to be zero. Define this macro for 7169targets where full C++ ABI compliance is impossible and where linker 7170restrictions require weak symbols to be left out of a static archive's 7171table of contents. 7172@end defmac 7173 7174@defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name}) 7175A C statement (sans semicolon) to output to the stdio stream 7176@var{stream} any text necessary for declaring the name of an external 7177symbol named @var{name} which is referenced in this compilation but 7178not defined. The value of @var{decl} is the tree node for the 7179declaration. 7180 7181This macro need not be defined if it does not need to output anything. 7182The GNU assembler and most Unix assemblers don't require anything. 7183@end defmac 7184 7185@deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref}) 7186This target hook is a function to output to @var{asm_out_file} an assembler 7187pseudo-op to declare a library function name external. The name of the 7188library function is given by @var{symref}, which is a @code{symbol_ref}. 7189@end deftypefn 7190 7191@deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (tree @var{decl}) 7192This target hook is a function to output to @var{asm_out_file} an assembler 7193directive to annotate used symbol. Darwin target use .no_dead_code_strip 7194directive. 7195@end deftypefn 7196 7197@defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name}) 7198A C statement (sans semicolon) to output to the stdio stream 7199@var{stream} a reference in assembler syntax to a label named 7200@var{name}. This should add @samp{_} to the front of the name, if that 7201is customary on your operating system, as it is in most Berkeley Unix 7202systems. This macro is used in @code{assemble_name}. 7203@end defmac 7204 7205@defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym}) 7206A C statement (sans semicolon) to output a reference to 7207@code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name} 7208will be used to output the name of the symbol. This macro may be used 7209to modify the way a symbol is referenced depending on information 7210encoded by @code{TARGET_ENCODE_SECTION_INFO}. 7211@end defmac 7212 7213@defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf}) 7214A C statement (sans semicolon) to output a reference to @var{buf}, the 7215result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined, 7216@code{assemble_name} will be used to output the name of the symbol. 7217This macro is not used by @code{output_asm_label}, or the @code{%l} 7218specifier that calls it; the intention is that this macro should be set 7219when it is necessary to output a label differently when its address is 7220being taken. 7221@end defmac 7222 7223@deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno}) 7224A function to output to the stdio stream @var{stream} a label whose 7225name is made from the string @var{prefix} and the number @var{labelno}. 7226 7227It is absolutely essential that these labels be distinct from the labels 7228used for user-level functions and variables. Otherwise, certain programs 7229will have name conflicts with internal labels. 7230 7231It is desirable to exclude internal labels from the symbol table of the 7232object file. Most assemblers have a naming convention for labels that 7233should be excluded; on many systems, the letter @samp{L} at the 7234beginning of a label has this effect. You should find out what 7235convention your system uses, and follow it. 7236 7237The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}. 7238@end deftypefn 7239 7240@defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num}) 7241A C statement to output to the stdio stream @var{stream} a debug info 7242label whose name is made from the string @var{prefix} and the number 7243@var{num}. This is useful for VLIW targets, where debug info labels 7244may need to be treated differently than branch target labels. On some 7245systems, branch target labels must be at the beginning of instruction 7246bundles, but debug info labels can occur in the middle of instruction 7247bundles. 7248 7249If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be 7250used. 7251@end defmac 7252 7253@defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num}) 7254A C statement to store into the string @var{string} a label whose name 7255is made from the string @var{prefix} and the number @var{num}. 7256 7257This string, when output subsequently by @code{assemble_name}, should 7258produce the output that @code{(*targetm.asm_out.internal_label)} would produce 7259with the same @var{prefix} and @var{num}. 7260 7261If the string begins with @samp{*}, then @code{assemble_name} will 7262output the rest of the string unchanged. It is often convenient for 7263@code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the 7264string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets 7265to output the string, and may change it. (Of course, 7266@code{ASM_OUTPUT_LABELREF} is also part of your machine description, so 7267you should know what it does on your machine.) 7268@end defmac 7269 7270@defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number}) 7271A C expression to assign to @var{outvar} (which is a variable of type 7272@code{char *}) a newly allocated string made from the string 7273@var{name} and the number @var{number}, with some suitable punctuation 7274added. Use @code{alloca} to get space for the string. 7275 7276The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to 7277produce an assembler label for an internal static variable whose name is 7278@var{name}. Therefore, the string must be such as to result in valid 7279assembler code. The argument @var{number} is different each time this 7280macro is executed; it prevents conflicts between similarly-named 7281internal static variables in different scopes. 7282 7283Ideally this string should not be a valid C identifier, to prevent any 7284conflict with the user's own symbols. Most assemblers allow periods 7285or percent signs in assembler symbols; putting at least one of these 7286between the name and the number will suffice. 7287 7288If this macro is not defined, a default definition will be provided 7289which is correct for most systems. 7290@end defmac 7291 7292@defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value}) 7293A C statement to output to the stdio stream @var{stream} assembler code 7294which defines (equates) the symbol @var{name} to have the value @var{value}. 7295 7296@findex SET_ASM_OP 7297If @code{SET_ASM_OP} is defined, a default definition is provided which is 7298correct for most systems. 7299@end defmac 7300 7301@defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value}) 7302A C statement to output to the stdio stream @var{stream} assembler code 7303which defines (equates) the symbol whose tree node is @var{decl_of_name} 7304to have the value of the tree node @var{decl_of_value}. This macro will 7305be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if 7306the tree nodes are available. 7307 7308@findex SET_ASM_OP 7309If @code{SET_ASM_OP} is defined, a default definition is provided which is 7310correct for most systems. 7311@end defmac 7312 7313@defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value}) 7314A C statement that evaluates to true if the assembler code which defines 7315(equates) the symbol whose tree node is @var{decl_of_name} to have the value 7316of the tree node @var{decl_of_value} should be emitted near the end of the 7317current compilation unit. The default is to not defer output of defines. 7318This macro affects defines output by @samp{ASM_OUTPUT_DEF} and 7319@samp{ASM_OUTPUT_DEF_FROM_DECLS}. 7320@end defmac 7321 7322@defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value}) 7323A C statement to output to the stdio stream @var{stream} assembler code 7324which defines (equates) the weak symbol @var{name} to have the value 7325@var{value}. If @var{value} is @code{NULL}, it defines @var{name} as 7326an undefined weak symbol. 7327 7328Define this macro if the target only supports weak aliases; define 7329@code{ASM_OUTPUT_DEF} instead if possible. 7330@end defmac 7331 7332@defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name}) 7333Define this macro to override the default assembler names used for 7334Objective-C methods. 7335 7336The default name is a unique method number followed by the name of the 7337class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of 7338the category is also included in the assembler name (e.g.@: 7339@samp{_1_Foo_Bar}). 7340 7341These names are safe on most systems, but make debugging difficult since 7342the method's selector is not present in the name. Therefore, particular 7343systems define other ways of computing names. 7344 7345@var{buf} is an expression of type @code{char *} which gives you a 7346buffer in which to store the name; its length is as long as 7347@var{class_name}, @var{cat_name} and @var{sel_name} put together, plus 734850 characters extra. 7349 7350The argument @var{is_inst} specifies whether the method is an instance 7351method or a class method; @var{class_name} is the name of the class; 7352@var{cat_name} is the name of the category (or @code{NULL} if the method is not 7353in a category); and @var{sel_name} is the name of the selector. 7354 7355On systems where the assembler can handle quoted names, you can use this 7356macro to provide more human-readable names. 7357@end defmac 7358 7359@defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name}) 7360A C statement (sans semicolon) to output to the stdio stream 7361@var{stream} commands to declare that the label @var{name} is an 7362Objective-C class reference. This is only needed for targets whose 7363linkers have special support for NeXT-style runtimes. 7364@end defmac 7365 7366@defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name}) 7367A C statement (sans semicolon) to output to the stdio stream 7368@var{stream} commands to declare that the label @var{name} is an 7369unresolved Objective-C class reference. This is only needed for targets 7370whose linkers have special support for NeXT-style runtimes. 7371@end defmac 7372 7373@node Initialization 7374@subsection How Initialization Functions Are Handled 7375@cindex initialization routines 7376@cindex termination routines 7377@cindex constructors, output of 7378@cindex destructors, output of 7379 7380The compiled code for certain languages includes @dfn{constructors} 7381(also called @dfn{initialization routines})---functions to initialize 7382data in the program when the program is started. These functions need 7383to be called before the program is ``started''---that is to say, before 7384@code{main} is called. 7385 7386Compiling some languages generates @dfn{destructors} (also called 7387@dfn{termination routines}) that should be called when the program 7388terminates. 7389 7390To make the initialization and termination functions work, the compiler 7391must output something in the assembler code to cause those functions to 7392be called at the appropriate time. When you port the compiler to a new 7393system, you need to specify how to do this. 7394 7395There are two major ways that GCC currently supports the execution of 7396initialization and termination functions. Each way has two variants. 7397Much of the structure is common to all four variations. 7398 7399@findex __CTOR_LIST__ 7400@findex __DTOR_LIST__ 7401The linker must build two lists of these functions---a list of 7402initialization functions, called @code{__CTOR_LIST__}, and a list of 7403termination functions, called @code{__DTOR_LIST__}. 7404 7405Each list always begins with an ignored function pointer (which may hold 74060, @minus{}1, or a count of the function pointers after it, depending on 7407the environment). This is followed by a series of zero or more function 7408pointers to constructors (or destructors), followed by a function 7409pointer containing zero. 7410 7411Depending on the operating system and its executable file format, either 7412@file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup 7413time and exit time. Constructors are called in reverse order of the 7414list; destructors in forward order. 7415 7416The best way to handle static constructors works only for object file 7417formats which provide arbitrarily-named sections. A section is set 7418aside for a list of constructors, and another for a list of destructors. 7419Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each 7420object file that defines an initialization function also puts a word in 7421the constructor section to point to that function. The linker 7422accumulates all these words into one contiguous @samp{.ctors} section. 7423Termination functions are handled similarly. 7424 7425This method will be chosen as the default by @file{target-def.h} if 7426@code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not 7427support arbitrary sections, but does support special designated 7428constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP} 7429and @code{DTORS_SECTION_ASM_OP} to achieve the same effect. 7430 7431When arbitrary sections are available, there are two variants, depending 7432upon how the code in @file{crtstuff.c} is called. On systems that 7433support a @dfn{.init} section which is executed at program startup, 7434parts of @file{crtstuff.c} are compiled into that section. The 7435program is linked by the @command{gcc} driver like this: 7436 7437@smallexample 7438ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o 7439@end smallexample 7440 7441The prologue of a function (@code{__init}) appears in the @code{.init} 7442section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise 7443for the function @code{__fini} in the @dfn{.fini} section. Normally these 7444files are provided by the operating system or by the GNU C library, but 7445are provided by GCC for a few targets. 7446 7447The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets) 7448compiled from @file{crtstuff.c}. They contain, among other things, code 7449fragments within the @code{.init} and @code{.fini} sections that branch 7450to routines in the @code{.text} section. The linker will pull all parts 7451of a section together, which results in a complete @code{__init} function 7452that invokes the routines we need at startup. 7453 7454To use this variant, you must define the @code{INIT_SECTION_ASM_OP} 7455macro properly. 7456 7457If no init section is available, when GCC compiles any function called 7458@code{main} (or more accurately, any function designated as a program 7459entry point by the language front end calling @code{expand_main_function}), 7460it inserts a procedure call to @code{__main} as the first executable code 7461after the function prologue. The @code{__main} function is defined 7462in @file{libgcc2.c} and runs the global constructors. 7463 7464In file formats that don't support arbitrary sections, there are again 7465two variants. In the simplest variant, the GNU linker (GNU @code{ld}) 7466and an `a.out' format must be used. In this case, 7467@code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs} 7468entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__}, 7469and with the address of the void function containing the initialization 7470code as its value. The GNU linker recognizes this as a request to add 7471the value to a @dfn{set}; the values are accumulated, and are eventually 7472placed in the executable as a vector in the format described above, with 7473a leading (ignored) count and a trailing zero element. 7474@code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init 7475section is available, the absence of @code{INIT_SECTION_ASM_OP} causes 7476the compilation of @code{main} to call @code{__main} as above, starting 7477the initialization process. 7478 7479The last variant uses neither arbitrary sections nor the GNU linker. 7480This is preferable when you want to do dynamic linking and when using 7481file formats which the GNU linker does not support, such as `ECOFF'@. In 7482this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and 7483termination functions are recognized simply by their names. This requires 7484an extra program in the linkage step, called @command{collect2}. This program 7485pretends to be the linker, for use with GCC; it does its job by running 7486the ordinary linker, but also arranges to include the vectors of 7487initialization and termination functions. These functions are called 7488via @code{__main} as described above. In order to use this method, 7489@code{use_collect2} must be defined in the target in @file{config.gcc}. 7490 7491@ifinfo 7492The following section describes the specific macros that control and 7493customize the handling of initialization and termination functions. 7494@end ifinfo 7495 7496@node Macros for Initialization 7497@subsection Macros Controlling Initialization Routines 7498 7499Here are the macros that control how the compiler handles initialization 7500and termination functions: 7501 7502@defmac INIT_SECTION_ASM_OP 7503If defined, a C string constant, including spacing, for the assembler 7504operation to identify the following data as initialization code. If not 7505defined, GCC will assume such a section does not exist. When you are 7506using special sections for initialization and termination functions, this 7507macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to 7508run the initialization functions. 7509@end defmac 7510 7511@defmac HAS_INIT_SECTION 7512If defined, @code{main} will not call @code{__main} as described above. 7513This macro should be defined for systems that control start-up code 7514on a symbol-by-symbol basis, such as OSF/1, and should not 7515be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}. 7516@end defmac 7517 7518@defmac LD_INIT_SWITCH 7519If defined, a C string constant for a switch that tells the linker that 7520the following symbol is an initialization routine. 7521@end defmac 7522 7523@defmac LD_FINI_SWITCH 7524If defined, a C string constant for a switch that tells the linker that 7525the following symbol is a finalization routine. 7526@end defmac 7527 7528@defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func}) 7529If defined, a C statement that will write a function that can be 7530automatically called when a shared library is loaded. The function 7531should call @var{func}, which takes no arguments. If not defined, and 7532the object format requires an explicit initialization function, then a 7533function called @code{_GLOBAL__DI} will be generated. 7534 7535This function and the following one are used by collect2 when linking a 7536shared library that needs constructors or destructors, or has DWARF2 7537exception tables embedded in the code. 7538@end defmac 7539 7540@defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func}) 7541If defined, a C statement that will write a function that can be 7542automatically called when a shared library is unloaded. The function 7543should call @var{func}, which takes no arguments. If not defined, and 7544the object format requires an explicit finalization function, then a 7545function called @code{_GLOBAL__DD} will be generated. 7546@end defmac 7547 7548@defmac INVOKE__main 7549If defined, @code{main} will call @code{__main} despite the presence of 7550@code{INIT_SECTION_ASM_OP}. This macro should be defined for systems 7551where the init section is not actually run automatically, but is still 7552useful for collecting the lists of constructors and destructors. 7553@end defmac 7554 7555@defmac SUPPORTS_INIT_PRIORITY 7556If nonzero, the C++ @code{init_priority} attribute is supported and the 7557compiler should emit instructions to control the order of initialization 7558of objects. If zero, the compiler will issue an error message upon 7559encountering an @code{init_priority} attribute. 7560@end defmac 7561 7562@deftypefn {Target Hook} bool TARGET_HAVE_CTORS_DTORS 7563This value is true if the target supports some ``native'' method of 7564collecting constructors and destructors to be run at startup and exit. 7565It is false if we must use @command{collect2}. 7566@end deftypefn 7567 7568@deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority}) 7569If defined, a function that outputs assembler code to arrange to call 7570the function referenced by @var{symbol} at initialization time. 7571 7572Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking 7573no arguments and with no return value. If the target supports initialization 7574priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY}; 7575otherwise it must be @code{DEFAULT_INIT_PRIORITY}. 7576 7577If this macro is not defined by the target, a suitable default will 7578be chosen if (1) the target supports arbitrary section names, (2) the 7579target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2} 7580is not defined. 7581@end deftypefn 7582 7583@deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority}) 7584This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination 7585functions rather than initialization functions. 7586@end deftypefn 7587 7588If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine 7589generated for the generated object file will have static linkage. 7590 7591If your system uses @command{collect2} as the means of processing 7592constructors, then that program normally uses @command{nm} to scan 7593an object file for constructor functions to be called. 7594 7595On certain kinds of systems, you can define this macro to make 7596@command{collect2} work faster (and, in some cases, make it work at all): 7597 7598@defmac OBJECT_FORMAT_COFF 7599Define this macro if the system uses COFF (Common Object File Format) 7600object files, so that @command{collect2} can assume this format and scan 7601object files directly for dynamic constructor/destructor functions. 7602 7603This macro is effective only in a native compiler; @command{collect2} as 7604part of a cross compiler always uses @command{nm} for the target machine. 7605@end defmac 7606 7607@defmac REAL_NM_FILE_NAME 7608Define this macro as a C string constant containing the file name to use 7609to execute @command{nm}. The default is to search the path normally for 7610@command{nm}. 7611 7612If your system supports shared libraries and has a program to list the 7613dynamic dependencies of a given library or executable, you can define 7614these macros to enable support for running initialization and 7615termination functions in shared libraries: 7616@end defmac 7617 7618@defmac LDD_SUFFIX 7619Define this macro to a C string constant containing the name of the program 7620which lists dynamic dependencies, like @command{"ldd"} under SunOS 4. 7621@end defmac 7622 7623@defmac PARSE_LDD_OUTPUT (@var{ptr}) 7624Define this macro to be C code that extracts filenames from the output 7625of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable 7626of type @code{char *} that points to the beginning of a line of output 7627from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the 7628code must advance @var{ptr} to the beginning of the filename on that 7629line. Otherwise, it must set @var{ptr} to @code{NULL}. 7630@end defmac 7631 7632@node Instruction Output 7633@subsection Output of Assembler Instructions 7634 7635@c prevent bad page break with this line 7636This describes assembler instruction output. 7637 7638@defmac REGISTER_NAMES 7639A C initializer containing the assembler's names for the machine 7640registers, each one as a C string constant. This is what translates 7641register numbers in the compiler into assembler language. 7642@end defmac 7643 7644@defmac ADDITIONAL_REGISTER_NAMES 7645If defined, a C initializer for an array of structures containing a name 7646and a register number. This macro defines additional names for hard 7647registers, thus allowing the @code{asm} option in declarations to refer 7648to registers using alternate names. 7649@end defmac 7650 7651@defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr}) 7652Define this macro if you are using an unusual assembler that 7653requires different names for the machine instructions. 7654 7655The definition is a C statement or statements which output an 7656assembler instruction opcode to the stdio stream @var{stream}. The 7657macro-operand @var{ptr} is a variable of type @code{char *} which 7658points to the opcode name in its ``internal'' form---the form that is 7659written in the machine description. The definition should output the 7660opcode name to @var{stream}, performing any translation you desire, and 7661increment the variable @var{ptr} to point at the end of the opcode 7662so that it will not be output twice. 7663 7664In fact, your macro definition may process less than the entire opcode 7665name, or more than the opcode name; but if you want to process text 7666that includes @samp{%}-sequences to substitute operands, you must take 7667care of the substitution yourself. Just be sure to increment 7668@var{ptr} over whatever text should not be output normally. 7669 7670@findex recog_data.operand 7671If you need to look at the operand values, they can be found as the 7672elements of @code{recog_data.operand}. 7673 7674If the macro definition does nothing, the instruction is output 7675in the usual way. 7676@end defmac 7677 7678@defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands}) 7679If defined, a C statement to be executed just prior to the output of 7680assembler code for @var{insn}, to modify the extracted operands so 7681they will be output differently. 7682 7683Here the argument @var{opvec} is the vector containing the operands 7684extracted from @var{insn}, and @var{noperands} is the number of 7685elements of the vector which contain meaningful data for this insn. 7686The contents of this vector are what will be used to convert the insn 7687template into assembler code, so you can change the assembler output 7688by changing the contents of the vector. 7689 7690This macro is useful when various assembler syntaxes share a single 7691file of instruction patterns; by defining this macro differently, you 7692can cause a large class of instructions to be output differently (such 7693as with rearranged operands). Naturally, variations in assembler 7694syntax affecting individual insn patterns ought to be handled by 7695writing conditional output routines in those patterns. 7696 7697If this macro is not defined, it is equivalent to a null statement. 7698@end defmac 7699 7700@defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code}) 7701A C compound statement to output to stdio stream @var{stream} the 7702assembler syntax for an instruction operand @var{x}. @var{x} is an 7703RTL expression. 7704 7705@var{code} is a value that can be used to specify one of several ways 7706of printing the operand. It is used when identical operands must be 7707printed differently depending on the context. @var{code} comes from 7708the @samp{%} specification that was used to request printing of the 7709operand. If the specification was just @samp{%@var{digit}} then 7710@var{code} is 0; if the specification was @samp{%@var{ltr} 7711@var{digit}} then @var{code} is the ASCII code for @var{ltr}. 7712 7713@findex reg_names 7714If @var{x} is a register, this macro should print the register's name. 7715The names can be found in an array @code{reg_names} whose type is 7716@code{char *[]}. @code{reg_names} is initialized from 7717@code{REGISTER_NAMES}. 7718 7719When the machine description has a specification @samp{%@var{punct}} 7720(a @samp{%} followed by a punctuation character), this macro is called 7721with a null pointer for @var{x} and the punctuation character for 7722@var{code}. 7723@end defmac 7724 7725@defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code}) 7726A C expression which evaluates to true if @var{code} is a valid 7727punctuation character for use in the @code{PRINT_OPERAND} macro. If 7728@code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no 7729punctuation characters (except for the standard one, @samp{%}) are used 7730in this way. 7731@end defmac 7732 7733@defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x}) 7734A C compound statement to output to stdio stream @var{stream} the 7735assembler syntax for an instruction operand that is a memory reference 7736whose address is @var{x}. @var{x} is an RTL expression. 7737 7738@cindex @code{TARGET_ENCODE_SECTION_INFO} usage 7739On some machines, the syntax for a symbolic address depends on the 7740section that the address refers to. On these machines, define the hook 7741@code{TARGET_ENCODE_SECTION_INFO} to store the information into the 7742@code{symbol_ref}, and then check for it here. @xref{Assembler 7743Format}. 7744@end defmac 7745 7746@findex dbr_sequence_length 7747@defmac DBR_OUTPUT_SEQEND (@var{file}) 7748A C statement, to be executed after all slot-filler instructions have 7749been output. If necessary, call @code{dbr_sequence_length} to 7750determine the number of slots filled in a sequence (zero if not 7751currently outputting a sequence), to decide how many no-ops to output, 7752or whatever. 7753 7754Don't define this macro if it has nothing to do, but it is helpful in 7755reading assembly output if the extent of the delay sequence is made 7756explicit (e.g.@: with white space). 7757@end defmac 7758 7759@findex final_sequence 7760Note that output routines for instructions with delay slots must be 7761prepared to deal with not being output as part of a sequence 7762(i.e.@: when the scheduling pass is not run, or when no slot fillers could be 7763found.) The variable @code{final_sequence} is null when not 7764processing a sequence, otherwise it contains the @code{sequence} rtx 7765being output. 7766 7767@findex asm_fprintf 7768@defmac REGISTER_PREFIX 7769@defmacx LOCAL_LABEL_PREFIX 7770@defmacx USER_LABEL_PREFIX 7771@defmacx IMMEDIATE_PREFIX 7772If defined, C string expressions to be used for the @samp{%R}, @samp{%L}, 7773@samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see 7774@file{final.c}). These are useful when a single @file{md} file must 7775support multiple assembler formats. In that case, the various @file{tm.h} 7776files can define these macros differently. 7777@end defmac 7778 7779@defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format}) 7780If defined this macro should expand to a series of @code{case} 7781statements which will be parsed inside the @code{switch} statement of 7782the @code{asm_fprintf} function. This allows targets to define extra 7783printf formats which may useful when generating their assembler 7784statements. Note that uppercase letters are reserved for future 7785generic extensions to asm_fprintf, and so are not available to target 7786specific code. The output file is given by the parameter @var{file}. 7787The varargs input pointer is @var{argptr} and the rest of the format 7788string, starting the character after the one that is being switched 7789upon, is pointed to by @var{format}. 7790@end defmac 7791 7792@defmac ASSEMBLER_DIALECT 7793If your target supports multiple dialects of assembler language (such as 7794different opcodes), define this macro as a C expression that gives the 7795numeric index of the assembler language dialect to use, with zero as the 7796first variant. 7797 7798If this macro is defined, you may use constructs of the form 7799@smallexample 7800@samp{@{option0|option1|option2@dots{}@}} 7801@end smallexample 7802@noindent 7803in the output templates of patterns (@pxref{Output Template}) or in the 7804first argument of @code{asm_fprintf}. This construct outputs 7805@samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of 7806@code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters 7807within these strings retain their usual meaning. If there are fewer 7808alternatives within the braces than the value of 7809@code{ASSEMBLER_DIALECT}, the construct outputs nothing. 7810 7811If you do not define this macro, the characters @samp{@{}, @samp{|} and 7812@samp{@}} do not have any special meaning when used in templates or 7813operands to @code{asm_fprintf}. 7814 7815Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX}, 7816@code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express 7817the variations in assembler language syntax with that mechanism. Define 7818@code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax 7819if the syntax variant are larger and involve such things as different 7820opcodes or operand order. 7821@end defmac 7822 7823@defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno}) 7824A C expression to output to @var{stream} some assembler code 7825which will push hard register number @var{regno} onto the stack. 7826The code need not be optimal, since this macro is used only when 7827profiling. 7828@end defmac 7829 7830@defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno}) 7831A C expression to output to @var{stream} some assembler code 7832which will pop hard register number @var{regno} off of the stack. 7833The code need not be optimal, since this macro is used only when 7834profiling. 7835@end defmac 7836 7837@node Dispatch Tables 7838@subsection Output of Dispatch Tables 7839 7840@c prevent bad page break with this line 7841This concerns dispatch tables. 7842 7843@cindex dispatch table 7844@defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel}) 7845A C statement to output to the stdio stream @var{stream} an assembler 7846pseudo-instruction to generate a difference between two labels. 7847@var{value} and @var{rel} are the numbers of two internal labels. The 7848definitions of these labels are output using 7849@code{(*targetm.asm_out.internal_label)}, and they must be printed in the same 7850way here. For example, 7851 7852@smallexample 7853fprintf (@var{stream}, "\t.word L%d-L%d\n", 7854 @var{value}, @var{rel}) 7855@end smallexample 7856 7857You must provide this macro on machines where the addresses in a 7858dispatch table are relative to the table's own address. If defined, GCC 7859will also use this macro on all machines when producing PIC@. 7860@var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the 7861mode and flags can be read. 7862@end defmac 7863 7864@defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value}) 7865This macro should be provided on machines where the addresses 7866in a dispatch table are absolute. 7867 7868The definition should be a C statement to output to the stdio stream 7869@var{stream} an assembler pseudo-instruction to generate a reference to 7870a label. @var{value} is the number of an internal label whose 7871definition is output using @code{(*targetm.asm_out.internal_label)}. 7872For example, 7873 7874@smallexample 7875fprintf (@var{stream}, "\t.word L%d\n", @var{value}) 7876@end smallexample 7877@end defmac 7878 7879@defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table}) 7880Define this if the label before a jump-table needs to be output 7881specially. The first three arguments are the same as for 7882@code{(*targetm.asm_out.internal_label)}; the fourth argument is the 7883jump-table which follows (a @code{jump_insn} containing an 7884@code{addr_vec} or @code{addr_diff_vec}). 7885 7886This feature is used on system V to output a @code{swbeg} statement 7887for the table. 7888 7889If this macro is not defined, these labels are output with 7890@code{(*targetm.asm_out.internal_label)}. 7891@end defmac 7892 7893@defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table}) 7894Define this if something special must be output at the end of a 7895jump-table. The definition should be a C statement to be executed 7896after the assembler code for the table is written. It should write 7897the appropriate code to stdio stream @var{stream}. The argument 7898@var{table} is the jump-table insn, and @var{num} is the label-number 7899of the preceding label. 7900 7901If this macro is not defined, nothing special is output at the end of 7902the jump-table. 7903@end defmac 7904 7905@deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (@var{stream}, @var{decl}, @var{for_eh}, @var{empty}) 7906This target hook emits a label at the beginning of each FDE@. It 7907should be defined on targets where FDEs need special labels, and it 7908should write the appropriate label, for the FDE associated with the 7909function declaration @var{decl}, to the stdio stream @var{stream}. 7910The third argument, @var{for_eh}, is a boolean: true if this is for an 7911exception table. The fourth argument, @var{empty}, is a boolean: 7912true if this is a placeholder label for an omitted FDE@. 7913 7914The default is that FDEs are not given nonlocal labels. 7915@end deftypefn 7916 7917@deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (@var{stream}) 7918This target hook emits a label at the beginning of the exception table. 7919It should be defined on targets where it is desirable for the table 7920to be broken up according to function. 7921 7922The default is that no label is emitted. 7923@end deftypefn 7924 7925@deftypefn {Target Hook} void TARGET_UNWIND_EMIT (FILE * @var{stream}, rtx @var{insn}) 7926This target hook emits and assembly directives required to unwind the 7927given instruction. This is only used when TARGET_UNWIND_INFO is set. 7928@end deftypefn 7929 7930@node Exception Region Output 7931@subsection Assembler Commands for Exception Regions 7932 7933@c prevent bad page break with this line 7934 7935This describes commands marking the start and the end of an exception 7936region. 7937 7938@defmac EH_FRAME_SECTION_NAME 7939If defined, a C string constant for the name of the section containing 7940exception handling frame unwind information. If not defined, GCC will 7941provide a default definition if the target supports named sections. 7942@file{crtstuff.c} uses this macro to switch to the appropriate section. 7943 7944You should define this symbol if your target supports DWARF 2 frame 7945unwind information and the default definition does not work. 7946@end defmac 7947 7948@defmac EH_FRAME_IN_DATA_SECTION 7949If defined, DWARF 2 frame unwind information will be placed in the 7950data section even though the target supports named sections. This 7951might be necessary, for instance, if the system linker does garbage 7952collection and sections cannot be marked as not to be collected. 7953 7954Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is 7955also defined. 7956@end defmac 7957 7958@defmac EH_TABLES_CAN_BE_READ_ONLY 7959Define this macro to 1 if your target is such that no frame unwind 7960information encoding used with non-PIC code will ever require a 7961runtime relocation, but the linker may not support merging read-only 7962and read-write sections into a single read-write section. 7963@end defmac 7964 7965@defmac MASK_RETURN_ADDR 7966An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so 7967that it does not contain any extraneous set bits in it. 7968@end defmac 7969 7970@defmac DWARF2_UNWIND_INFO 7971Define this macro to 0 if your target supports DWARF 2 frame unwind 7972information, but it does not yet work with exception handling. 7973Otherwise, if your target supports this information (if it defines 7974@samp{INCOMING_RETURN_ADDR_RTX} and either @samp{UNALIGNED_INT_ASM_OP} 7975or @samp{OBJECT_FORMAT_ELF}), GCC will provide a default definition of 1. 7976 7977If @code{TARGET_UNWIND_INFO} is defined, the target specific unwinder 7978will be used in all cases. Defining this macro will enable the generation 7979of DWARF 2 frame debugging information. 7980 7981If @code{TARGET_UNWIND_INFO} is not defined, and this macro is defined to 1, 7982the DWARF 2 unwinder will be the default exception handling mechanism; 7983otherwise, the @code{setjmp}/@code{longjmp}-based scheme will be used by 7984default. 7985@end defmac 7986 7987@defmac TARGET_UNWIND_INFO 7988Define this macro if your target has ABI specified unwind tables. Usually 7989these will be output by @code{TARGET_UNWIND_EMIT}. 7990@end defmac 7991 7992@deftypevar {Target Hook} bool TARGET_UNWIND_TABLES_DEFAULT 7993This variable should be set to @code{true} if the target ABI requires unwinding 7994tables even when exceptions are not used. 7995@end deftypevar 7996 7997@defmac MUST_USE_SJLJ_EXCEPTIONS 7998This macro need only be defined if @code{DWARF2_UNWIND_INFO} is 7999runtime-variable. In that case, @file{except.h} cannot correctly 8000determine the corresponding definition of @code{MUST_USE_SJLJ_EXCEPTIONS}, 8001so the target must provide it directly. 8002@end defmac 8003 8004@defmac DONT_USE_BUILTIN_SETJMP 8005Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme 8006should use the @code{setjmp}/@code{longjmp} functions from the C library 8007instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery. 8008@end defmac 8009 8010@defmac DWARF_CIE_DATA_ALIGNMENT 8011This macro need only be defined if the target might save registers in the 8012function prologue at an offset to the stack pointer that is not aligned to 8013@code{UNITS_PER_WORD}. The definition should be the negative minimum 8014alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive 8015minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if 8016the target supports DWARF 2 frame unwind information. 8017@end defmac 8018 8019@deftypevar {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO 8020Contains the value true if the target should add a zero word onto the 8021end of a Dwarf-2 frame info section when used for exception handling. 8022Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and 8023true otherwise. 8024@end deftypevar 8025 8026@deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg}) 8027Given a register, this hook should return a parallel of registers to 8028represent where to find the register pieces. Define this hook if the 8029register and its mode are represented in Dwarf in non-contiguous 8030locations, or if the register should be represented in more than one 8031register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}. 8032If not defined, the default is to return @code{NULL_RTX}. 8033@end deftypefn 8034 8035@deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym}) 8036This hook is used to output a reference from a frame unwinding table to 8037the type_info object identified by @var{sym}. It should return @code{true} 8038if the reference was output. Returning @code{false} will cause the 8039reference to be output using the normal Dwarf2 routines. 8040@end deftypefn 8041 8042@deftypefn {Target Hook} bool TARGET_ARM_EABI_UNWINDER 8043This hook should be set to @code{true} on targets that use an ARM EABI 8044based unwinding library, and @code{false} on other targets. This effects 8045the format of unwinding tables, and how the unwinder in entered after 8046running a cleanup. The default is @code{false}. 8047@end deftypefn 8048 8049@node Alignment Output 8050@subsection Assembler Commands for Alignment 8051 8052@c prevent bad page break with this line 8053This describes commands for alignment. 8054 8055@defmac JUMP_ALIGN (@var{label}) 8056The alignment (log base 2) to put in front of @var{label}, which is 8057a common destination of jumps and has no fallthru incoming edge. 8058 8059This macro need not be defined if you don't want any special alignment 8060to be done at such a time. Most machine descriptions do not currently 8061define the macro. 8062 8063Unless it's necessary to inspect the @var{label} parameter, it is better 8064to set the variable @var{align_jumps} in the target's 8065@code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's 8066selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation. 8067@end defmac 8068 8069@defmac LABEL_ALIGN_AFTER_BARRIER (@var{label}) 8070The alignment (log base 2) to put in front of @var{label}, which follows 8071a @code{BARRIER}. 8072 8073This macro need not be defined if you don't want any special alignment 8074to be done at such a time. Most machine descriptions do not currently 8075define the macro. 8076@end defmac 8077 8078@defmac LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP 8079The maximum number of bytes to skip when applying 8080@code{LABEL_ALIGN_AFTER_BARRIER}. This works only if 8081@code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 8082@end defmac 8083 8084@defmac LOOP_ALIGN (@var{label}) 8085The alignment (log base 2) to put in front of @var{label}, which follows 8086a @code{NOTE_INSN_LOOP_BEG} note. 8087 8088This macro need not be defined if you don't want any special alignment 8089to be done at such a time. Most machine descriptions do not currently 8090define the macro. 8091 8092Unless it's necessary to inspect the @var{label} parameter, it is better 8093to set the variable @code{align_loops} in the target's 8094@code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's 8095selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation. 8096@end defmac 8097 8098@defmac LOOP_ALIGN_MAX_SKIP 8099The maximum number of bytes to skip when applying @code{LOOP_ALIGN}. 8100This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 8101@end defmac 8102 8103@defmac LABEL_ALIGN (@var{label}) 8104The alignment (log base 2) to put in front of @var{label}. 8105If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment, 8106the maximum of the specified values is used. 8107 8108Unless it's necessary to inspect the @var{label} parameter, it is better 8109to set the variable @code{align_labels} in the target's 8110@code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's 8111selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation. 8112@end defmac 8113 8114@defmac LABEL_ALIGN_MAX_SKIP 8115The maximum number of bytes to skip when applying @code{LABEL_ALIGN}. 8116This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 8117@end defmac 8118 8119@defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes}) 8120A C statement to output to the stdio stream @var{stream} an assembler 8121instruction to advance the location counter by @var{nbytes} bytes. 8122Those bytes should be zero when loaded. @var{nbytes} will be a C 8123expression of type @code{int}. 8124@end defmac 8125 8126@defmac ASM_NO_SKIP_IN_TEXT 8127Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the 8128text section because it fails to put zeros in the bytes that are skipped. 8129This is true on many Unix systems, where the pseudo--op to skip bytes 8130produces no-op instructions rather than zeros when used in the text 8131section. 8132@end defmac 8133 8134@defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power}) 8135A C statement to output to the stdio stream @var{stream} an assembler 8136command to advance the location counter to a multiple of 2 to the 8137@var{power} bytes. @var{power} will be a C expression of type @code{int}. 8138@end defmac 8139 8140@defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power}) 8141Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used 8142for padding, if necessary. 8143@end defmac 8144 8145@defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip}) 8146A C statement to output to the stdio stream @var{stream} an assembler 8147command to advance the location counter to a multiple of 2 to the 8148@var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to 8149satisfy the alignment request. @var{power} and @var{max_skip} will be 8150a C expression of type @code{int}. 8151@end defmac 8152 8153@need 3000 8154@node Debugging Info 8155@section Controlling Debugging Information Format 8156 8157@c prevent bad page break with this line 8158This describes how to specify debugging information. 8159 8160@menu 8161* All Debuggers:: Macros that affect all debugging formats uniformly. 8162* DBX Options:: Macros enabling specific options in DBX format. 8163* DBX Hooks:: Hook macros for varying DBX format. 8164* File Names and DBX:: Macros controlling output of file names in DBX format. 8165* SDB and DWARF:: Macros for SDB (COFF) and DWARF formats. 8166* VMS Debug:: Macros for VMS debug format. 8167@end menu 8168 8169@node All Debuggers 8170@subsection Macros Affecting All Debugging Formats 8171 8172@c prevent bad page break with this line 8173These macros affect all debugging formats. 8174 8175@defmac DBX_REGISTER_NUMBER (@var{regno}) 8176A C expression that returns the DBX register number for the compiler 8177register number @var{regno}. In the default macro provided, the value 8178of this expression will be @var{regno} itself. But sometimes there are 8179some registers that the compiler knows about and DBX does not, or vice 8180versa. In such cases, some register may need to have one number in the 8181compiler and another for DBX@. 8182 8183If two registers have consecutive numbers inside GCC, and they can be 8184used as a pair to hold a multiword value, then they @emph{must} have 8185consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}. 8186Otherwise, debuggers will be unable to access such a pair, because they 8187expect register pairs to be consecutive in their own numbering scheme. 8188 8189If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that 8190does not preserve register pairs, then what you must do instead is 8191redefine the actual register numbering scheme. 8192@end defmac 8193 8194@defmac DEBUGGER_AUTO_OFFSET (@var{x}) 8195A C expression that returns the integer offset value for an automatic 8196variable having address @var{x} (an RTL expression). The default 8197computation assumes that @var{x} is based on the frame-pointer and 8198gives the offset from the frame-pointer. This is required for targets 8199that produce debugging output for DBX or COFF-style debugging output 8200for SDB and allow the frame-pointer to be eliminated when the 8201@option{-g} options is used. 8202@end defmac 8203 8204@defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x}) 8205A C expression that returns the integer offset value for an argument 8206having address @var{x} (an RTL expression). The nominal offset is 8207@var{offset}. 8208@end defmac 8209 8210@defmac PREFERRED_DEBUGGING_TYPE 8211A C expression that returns the type of debugging output GCC should 8212produce when the user specifies just @option{-g}. Define 8213this if you have arranged for GCC to support more than one format of 8214debugging output. Currently, the allowable values are @code{DBX_DEBUG}, 8215@code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG}, 8216@code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}. 8217 8218When the user specifies @option{-ggdb}, GCC normally also uses the 8219value of this macro to select the debugging output format, but with two 8220exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the 8221value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is 8222defined, GCC uses @code{DBX_DEBUG}. 8223 8224The value of this macro only affects the default debugging output; the 8225user can always get a specific type of output by using @option{-gstabs}, 8226@option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}. 8227@end defmac 8228 8229@node DBX Options 8230@subsection Specific Options for DBX Output 8231 8232@c prevent bad page break with this line 8233These are specific options for DBX output. 8234 8235@defmac DBX_DEBUGGING_INFO 8236Define this macro if GCC should produce debugging output for DBX 8237in response to the @option{-g} option. 8238@end defmac 8239 8240@defmac XCOFF_DEBUGGING_INFO 8241Define this macro if GCC should produce XCOFF format debugging output 8242in response to the @option{-g} option. This is a variant of DBX format. 8243@end defmac 8244 8245@defmac DEFAULT_GDB_EXTENSIONS 8246Define this macro to control whether GCC should by default generate 8247GDB's extended version of DBX debugging information (assuming DBX-format 8248debugging information is enabled at all). If you don't define the 8249macro, the default is 1: always generate the extended information 8250if there is any occasion to. 8251@end defmac 8252 8253@defmac DEBUG_SYMS_TEXT 8254Define this macro if all @code{.stabs} commands should be output while 8255in the text section. 8256@end defmac 8257 8258@defmac ASM_STABS_OP 8259A C string constant, including spacing, naming the assembler pseudo op to 8260use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol. 8261If you don't define this macro, @code{"\t.stabs\t"} is used. This macro 8262applies only to DBX debugging information format. 8263@end defmac 8264 8265@defmac ASM_STABD_OP 8266A C string constant, including spacing, naming the assembler pseudo op to 8267use instead of @code{"\t.stabd\t"} to define a debugging symbol whose 8268value is the current location. If you don't define this macro, 8269@code{"\t.stabd\t"} is used. This macro applies only to DBX debugging 8270information format. 8271@end defmac 8272 8273@defmac ASM_STABN_OP 8274A C string constant, including spacing, naming the assembler pseudo op to 8275use instead of @code{"\t.stabn\t"} to define a debugging symbol with no 8276name. If you don't define this macro, @code{"\t.stabn\t"} is used. This 8277macro applies only to DBX debugging information format. 8278@end defmac 8279 8280@defmac DBX_NO_XREFS 8281Define this macro if DBX on your system does not support the construct 8282@samp{xs@var{tagname}}. On some systems, this construct is used to 8283describe a forward reference to a structure named @var{tagname}. 8284On other systems, this construct is not supported at all. 8285@end defmac 8286 8287@defmac DBX_CONTIN_LENGTH 8288A symbol name in DBX-format debugging information is normally 8289continued (split into two separate @code{.stabs} directives) when it 8290exceeds a certain length (by default, 80 characters). On some 8291operating systems, DBX requires this splitting; on others, splitting 8292must not be done. You can inhibit splitting by defining this macro 8293with the value zero. You can override the default splitting-length by 8294defining this macro as an expression for the length you desire. 8295@end defmac 8296 8297@defmac DBX_CONTIN_CHAR 8298Normally continuation is indicated by adding a @samp{\} character to 8299the end of a @code{.stabs} string when a continuation follows. To use 8300a different character instead, define this macro as a character 8301constant for the character you want to use. Do not define this macro 8302if backslash is correct for your system. 8303@end defmac 8304 8305@defmac DBX_STATIC_STAB_DATA_SECTION 8306Define this macro if it is necessary to go to the data section before 8307outputting the @samp{.stabs} pseudo-op for a non-global static 8308variable. 8309@end defmac 8310 8311@defmac DBX_TYPE_DECL_STABS_CODE 8312The value to use in the ``code'' field of the @code{.stabs} directive 8313for a typedef. The default is @code{N_LSYM}. 8314@end defmac 8315 8316@defmac DBX_STATIC_CONST_VAR_CODE 8317The value to use in the ``code'' field of the @code{.stabs} directive 8318for a static variable located in the text section. DBX format does not 8319provide any ``right'' way to do this. The default is @code{N_FUN}. 8320@end defmac 8321 8322@defmac DBX_REGPARM_STABS_CODE 8323The value to use in the ``code'' field of the @code{.stabs} directive 8324for a parameter passed in registers. DBX format does not provide any 8325``right'' way to do this. The default is @code{N_RSYM}. 8326@end defmac 8327 8328@defmac DBX_REGPARM_STABS_LETTER 8329The letter to use in DBX symbol data to identify a symbol as a parameter 8330passed in registers. DBX format does not customarily provide any way to 8331do this. The default is @code{'P'}. 8332@end defmac 8333 8334@defmac DBX_FUNCTION_FIRST 8335Define this macro if the DBX information for a function and its 8336arguments should precede the assembler code for the function. Normally, 8337in DBX format, the debugging information entirely follows the assembler 8338code. 8339@end defmac 8340 8341@defmac DBX_BLOCKS_FUNCTION_RELATIVE 8342Define this macro, with value 1, if the value of a symbol describing 8343the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be 8344relative to the start of the enclosing function. Normally, GCC uses 8345an absolute address. 8346@end defmac 8347 8348@defmac DBX_LINES_FUNCTION_RELATIVE 8349Define this macro, with value 1, if the value of a symbol indicating 8350the current line number (@code{N_SLINE}) should be relative to the 8351start of the enclosing function. Normally, GCC uses an absolute address. 8352@end defmac 8353 8354@defmac DBX_USE_BINCL 8355Define this macro if GCC should generate @code{N_BINCL} and 8356@code{N_EINCL} stabs for included header files, as on Sun systems. This 8357macro also directs GCC to output a type number as a pair of a file 8358number and a type number within the file. Normally, GCC does not 8359generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single 8360number for a type number. 8361@end defmac 8362 8363@node DBX Hooks 8364@subsection Open-Ended Hooks for DBX Format 8365 8366@c prevent bad page break with this line 8367These are hooks for DBX format. 8368 8369@defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name}) 8370Define this macro to say how to output to @var{stream} the debugging 8371information for the start of a scope level for variable names. The 8372argument @var{name} is the name of an assembler symbol (for use with 8373@code{assemble_name}) whose value is the address where the scope begins. 8374@end defmac 8375 8376@defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name}) 8377Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level. 8378@end defmac 8379 8380@defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl}) 8381Define this macro if the target machine requires special handling to 8382output an @code{N_FUN} entry for the function @var{decl}. 8383@end defmac 8384 8385@defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter}) 8386A C statement to output DBX debugging information before code for line 8387number @var{line} of the current source file to the stdio stream 8388@var{stream}. @var{counter} is the number of time the macro was 8389invoked, including the current invocation; it is intended to generate 8390unique labels in the assembly output. 8391 8392This macro should not be defined if the default output is correct, or 8393if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}. 8394@end defmac 8395 8396@defmac NO_DBX_FUNCTION_END 8397Some stabs encapsulation formats (in particular ECOFF), cannot handle the 8398@code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct. 8399On those machines, define this macro to turn this feature off without 8400disturbing the rest of the gdb extensions. 8401@end defmac 8402 8403@defmac NO_DBX_BNSYM_ENSYM 8404Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx 8405extension construct. On those machines, define this macro to turn this 8406feature off without disturbing the rest of the gdb extensions. 8407@end defmac 8408 8409@node File Names and DBX 8410@subsection File Names in DBX Format 8411 8412@c prevent bad page break with this line 8413This describes file names in DBX format. 8414 8415@defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name}) 8416A C statement to output DBX debugging information to the stdio stream 8417@var{stream}, which indicates that file @var{name} is the main source 8418file---the file specified as the input file for compilation. 8419This macro is called only once, at the beginning of compilation. 8420 8421This macro need not be defined if the standard form of output 8422for DBX debugging information is appropriate. 8423 8424It may be necessary to refer to a label equal to the beginning of the 8425text section. You can use @samp{assemble_name (stream, ltext_label_name)} 8426to do so. If you do this, you must also set the variable 8427@var{used_ltext_label_name} to @code{true}. 8428@end defmac 8429 8430@defmac NO_DBX_MAIN_SOURCE_DIRECTORY 8431Define this macro, with value 1, if GCC should not emit an indication 8432of the current directory for compilation and current source language at 8433the beginning of the file. 8434@end defmac 8435 8436@defmac NO_DBX_GCC_MARKER 8437Define this macro, with value 1, if GCC should not emit an indication 8438that this object file was compiled by GCC@. The default is to emit 8439an @code{N_OPT} stab at the beginning of every source file, with 8440@samp{gcc2_compiled.} for the string and value 0. 8441@end defmac 8442 8443@defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name}) 8444A C statement to output DBX debugging information at the end of 8445compilation of the main source file @var{name}. Output should be 8446written to the stdio stream @var{stream}. 8447 8448If you don't define this macro, nothing special is output at the end 8449of compilation, which is correct for most machines. 8450@end defmac 8451 8452@defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END 8453Define this macro @emph{instead of} defining 8454@code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at 8455the end of compilation is a @code{N_SO} stab with an empty string, 8456whose value is the highest absolute text address in the file. 8457@end defmac 8458 8459@need 2000 8460@node SDB and DWARF 8461@subsection Macros for SDB and DWARF Output 8462 8463@c prevent bad page break with this line 8464Here are macros for SDB and DWARF output. 8465 8466@defmac SDB_DEBUGGING_INFO 8467Define this macro if GCC should produce COFF-style debugging output 8468for SDB in response to the @option{-g} option. 8469@end defmac 8470 8471@defmac DWARF2_DEBUGGING_INFO 8472Define this macro if GCC should produce dwarf version 2 format 8473debugging output in response to the @option{-g} option. 8474 8475@deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (tree @var{function}) 8476Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to 8477be emitted for each function. Instead of an integer return the enum 8478value for the @code{DW_CC_} tag. 8479@end deftypefn 8480 8481To support optional call frame debugging information, you must also 8482define @code{INCOMING_RETURN_ADDR_RTX} and either set 8483@code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the 8484prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save} 8485as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't. 8486@end defmac 8487 8488@defmac DWARF2_FRAME_INFO 8489Define this macro to a nonzero value if GCC should always output 8490Dwarf 2 frame information. If @code{DWARF2_UNWIND_INFO} 8491(@pxref{Exception Region Output} is nonzero, GCC will output this 8492information not matter how you define @code{DWARF2_FRAME_INFO}. 8493@end defmac 8494 8495@defmac DWARF2_ASM_LINE_DEBUG_INFO 8496Define this macro to be a nonzero value if the assembler can generate Dwarf 2 8497line debug info sections. This will result in much more compact line number 8498tables, and hence is desirable if it works. 8499@end defmac 8500 8501@defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2}) 8502A C statement to issue assembly directives that create a difference 8503@var{lab1} minus @var{lab2}, using an integer of the given @var{size}. 8504@end defmac 8505 8506@defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section}) 8507A C statement to issue assembly directives that create a 8508section-relative reference to the given @var{label}, using an integer of the 8509given @var{size}. The label is known to be defined in the given @var{section}. 8510@end defmac 8511 8512@defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label}) 8513A C statement to issue assembly directives that create a self-relative 8514reference to the given @var{label}, using an integer of the given @var{size}. 8515@end defmac 8516 8517@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{FILE}, int @var{size}, rtx @var{x}) 8518If defined, this target hook is a function which outputs a DTP-relative 8519reference to the given TLS symbol of the specified size. 8520@end deftypefn 8521 8522@defmac PUT_SDB_@dots{} 8523Define these macros to override the assembler syntax for the special 8524SDB assembler directives. See @file{sdbout.c} for a list of these 8525macros and their arguments. If the standard syntax is used, you need 8526not define them yourself. 8527@end defmac 8528 8529@defmac SDB_DELIM 8530Some assemblers do not support a semicolon as a delimiter, even between 8531SDB assembler directives. In that case, define this macro to be the 8532delimiter to use (usually @samp{\n}). It is not necessary to define 8533a new set of @code{PUT_SDB_@var{op}} macros if this is the only change 8534required. 8535@end defmac 8536 8537@defmac SDB_ALLOW_UNKNOWN_REFERENCES 8538Define this macro to allow references to unknown structure, 8539union, or enumeration tags to be emitted. Standard COFF does not 8540allow handling of unknown references, MIPS ECOFF has support for 8541it. 8542@end defmac 8543 8544@defmac SDB_ALLOW_FORWARD_REFERENCES 8545Define this macro to allow references to structure, union, or 8546enumeration tags that have not yet been seen to be handled. Some 8547assemblers choke if forward tags are used, while some require it. 8548@end defmac 8549 8550@defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}) 8551A C statement to output SDB debugging information before code for line 8552number @var{line} of the current source file to the stdio stream 8553@var{stream}. The default is to emit an @code{.ln} directive. 8554@end defmac 8555 8556@need 2000 8557@node VMS Debug 8558@subsection Macros for VMS Debug Format 8559 8560@c prevent bad page break with this line 8561Here are macros for VMS debug format. 8562 8563@defmac VMS_DEBUGGING_INFO 8564Define this macro if GCC should produce debugging output for VMS 8565in response to the @option{-g} option. The default behavior for VMS 8566is to generate minimal debug info for a traceback in the absence of 8567@option{-g} unless explicitly overridden with @option{-g0}. This 8568behavior is controlled by @code{OPTIMIZATION_OPTIONS} and 8569@code{OVERRIDE_OPTIONS}. 8570@end defmac 8571 8572@node Floating Point 8573@section Cross Compilation and Floating Point 8574@cindex cross compilation and floating point 8575@cindex floating point and cross compilation 8576 8577While all modern machines use twos-complement representation for integers, 8578there are a variety of representations for floating point numbers. This 8579means that in a cross-compiler the representation of floating point numbers 8580in the compiled program may be different from that used in the machine 8581doing the compilation. 8582 8583Because different representation systems may offer different amounts of 8584range and precision, all floating point constants must be represented in 8585the target machine's format. Therefore, the cross compiler cannot 8586safely use the host machine's floating point arithmetic; it must emulate 8587the target's arithmetic. To ensure consistency, GCC always uses 8588emulation to work with floating point values, even when the host and 8589target floating point formats are identical. 8590 8591The following macros are provided by @file{real.h} for the compiler to 8592use. All parts of the compiler which generate or optimize 8593floating-point calculations must use these macros. They may evaluate 8594their operands more than once, so operands must not have side effects. 8595 8596@defmac REAL_VALUE_TYPE 8597The C data type to be used to hold a floating point value in the target 8598machine's format. Typically this is a @code{struct} containing an 8599array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque 8600quantity. 8601@end defmac 8602 8603@deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y}) 8604Compares for equality the two values, @var{x} and @var{y}. If the target 8605floating point format supports negative zeroes and/or NaNs, 8606@samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and 8607@samp{REAL_VALUES_EQUAL (NaN, NaN)} is false. 8608@end deftypefn 8609 8610@deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y}) 8611Tests whether @var{x} is less than @var{y}. 8612@end deftypefn 8613 8614@deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x}) 8615Truncates @var{x} to a signed integer, rounding toward zero. 8616@end deftypefn 8617 8618@deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x}) 8619Truncates @var{x} to an unsigned integer, rounding toward zero. If 8620@var{x} is negative, returns zero. 8621@end deftypefn 8622 8623@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode}) 8624Converts @var{string} into a floating point number in the target machine's 8625representation for mode @var{mode}. This routine can handle both 8626decimal and hexadecimal floating point constants, using the syntax 8627defined by the C language for both. 8628@end deftypefn 8629 8630@deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x}) 8631Returns 1 if @var{x} is negative (including negative zero), 0 otherwise. 8632@end deftypefn 8633 8634@deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x}) 8635Determines whether @var{x} represents infinity (positive or negative). 8636@end deftypefn 8637 8638@deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x}) 8639Determines whether @var{x} represents a ``NaN'' (not-a-number). 8640@end deftypefn 8641 8642@deftypefn Macro void REAL_ARITHMETIC (REAL_VALUE_TYPE @var{output}, enum tree_code @var{code}, REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y}) 8643Calculates an arithmetic operation on the two floating point values 8644@var{x} and @var{y}, storing the result in @var{output} (which must be a 8645variable). 8646 8647The operation to be performed is specified by @var{code}. Only the 8648following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR}, 8649@code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}. 8650 8651If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the 8652target's floating point format cannot represent infinity, it will call 8653@code{abort}. Callers should check for this situation first, using 8654@code{MODE_HAS_INFINITIES}. @xref{Storage Layout}. 8655@end deftypefn 8656 8657@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x}) 8658Returns the negative of the floating point value @var{x}. 8659@end deftypefn 8660 8661@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x}) 8662Returns the absolute value of @var{x}. 8663@end deftypefn 8664 8665@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x}) 8666Truncates the floating point value @var{x} to fit in @var{mode}. The 8667return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an 8668appropriate bit pattern to be output asa floating constant whose 8669precision accords with mode @var{mode}. 8670@end deftypefn 8671 8672@deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x}) 8673Converts a floating point value @var{x} into a double-precision integer 8674which is then stored into @var{low} and @var{high}. If the value is not 8675integral, it is truncated. 8676@end deftypefn 8677 8678@deftypefn Macro void REAL_VALUE_FROM_INT (REAL_VALUE_TYPE @var{x}, HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, enum machine_mode @var{mode}) 8679Converts a double-precision integer found in @var{low} and @var{high}, 8680into a floating point value which is then stored into @var{x}. The 8681value is truncated to fit in mode @var{mode}. 8682@end deftypefn 8683 8684@node Mode Switching 8685@section Mode Switching Instructions 8686@cindex mode switching 8687The following macros control mode switching optimizations: 8688 8689@defmac OPTIMIZE_MODE_SWITCHING (@var{entity}) 8690Define this macro if the port needs extra instructions inserted for mode 8691switching in an optimizing compilation. 8692 8693For an example, the SH4 can perform both single and double precision 8694floating point operations, but to perform a single precision operation, 8695the FPSCR PR bit has to be cleared, while for a double precision 8696operation, this bit has to be set. Changing the PR bit requires a general 8697purpose register as a scratch register, hence these FPSCR sets have to 8698be inserted before reload, i.e.@: you can't put this into instruction emitting 8699or @code{TARGET_MACHINE_DEPENDENT_REORG}. 8700 8701You can have multiple entities that are mode-switched, and select at run time 8702which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should 8703return nonzero for any @var{entity} that needs mode-switching. 8704If you define this macro, you also have to define 8705@code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED}, 8706@code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}. 8707@code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT} 8708are optional. 8709@end defmac 8710 8711@defmac NUM_MODES_FOR_MODE_SWITCHING 8712If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as 8713initializer for an array of integers. Each initializer element 8714N refers to an entity that needs mode switching, and specifies the number 8715of different modes that might need to be set for this entity. 8716The position of the initializer in the initializer---starting counting at 8717zero---determines the integer that is used to refer to the mode-switched 8718entity in question. 8719In macros that take mode arguments / yield a mode result, modes are 8720represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode 8721switch is needed / supplied. 8722@end defmac 8723 8724@defmac MODE_NEEDED (@var{entity}, @var{insn}) 8725@var{entity} is an integer specifying a mode-switched entity. If 8726@code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to 8727return an integer value not larger than the corresponding element in 8728@code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must 8729be switched into prior to the execution of @var{insn}. 8730@end defmac 8731 8732@defmac MODE_AFTER (@var{mode}, @var{insn}) 8733If this macro is defined, it is evaluated for every @var{insn} during 8734mode switching. It determines the mode that an insn results in (if 8735different from the incoming mode). 8736@end defmac 8737 8738@defmac MODE_ENTRY (@var{entity}) 8739If this macro is defined, it is evaluated for every @var{entity} that needs 8740mode switching. It should evaluate to an integer, which is a mode that 8741@var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY} 8742is defined then @code{MODE_EXIT} must be defined. 8743@end defmac 8744 8745@defmac MODE_EXIT (@var{entity}) 8746If this macro is defined, it is evaluated for every @var{entity} that needs 8747mode switching. It should evaluate to an integer, which is a mode that 8748@var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT} 8749is defined then @code{MODE_ENTRY} must be defined. 8750@end defmac 8751 8752@defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n}) 8753This macro specifies the order in which modes for @var{entity} are processed. 87540 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the 8755lowest. The value of the macro should be an integer designating a mode 8756for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode} 8757(@var{entity}, @var{n}) shall be a bijection in 0 @dots{} 8758@code{num_modes_for_mode_switching[@var{entity}] - 1}. 8759@end defmac 8760 8761@defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live}) 8762Generate one or more insns to set @var{entity} to @var{mode}. 8763@var{hard_reg_live} is the set of hard registers live at the point where 8764the insn(s) are to be inserted. 8765@end defmac 8766 8767@node Target Attributes 8768@section Defining target-specific uses of @code{__attribute__} 8769@cindex target attributes 8770@cindex machine attributes 8771@cindex attributes, target-specific 8772 8773Target-specific attributes may be defined for functions, data and types. 8774These are described using the following target hooks; they also need to 8775be documented in @file{extend.texi}. 8776 8777@deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE 8778If defined, this target hook points to an array of @samp{struct 8779attribute_spec} (defined in @file{tree.h}) specifying the machine 8780specific attributes for this target and some of the restrictions on the 8781entities to which these attributes are applied and the arguments they 8782take. 8783@end deftypevr 8784 8785@deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2}) 8786If defined, this target hook is a function which returns zero if the attributes on 8787@var{type1} and @var{type2} are incompatible, one if they are compatible, 8788and two if they are nearly compatible (which causes a warning to be 8789generated). If this is not defined, machine-specific attributes are 8790supposed always to be compatible. 8791@end deftypefn 8792 8793@deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type}) 8794If defined, this target hook is a function which assigns default attributes to 8795newly defined @var{type}. 8796@end deftypefn 8797 8798@deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2}) 8799Define this target hook if the merging of type attributes needs special 8800handling. If defined, the result is a list of the combined 8801@code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed 8802that @code{comptypes} has already been called and returned 1. This 8803function may call @code{merge_attributes} to handle machine-independent 8804merging. 8805@end deftypefn 8806 8807@deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl}) 8808Define this target hook if the merging of decl attributes needs special 8809handling. If defined, the result is a list of the combined 8810@code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}. 8811@var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of 8812when this is needed are when one attribute overrides another, or when an 8813attribute is nullified by a subsequent definition. This function may 8814call @code{merge_attributes} to handle machine-independent merging. 8815 8816@findex TARGET_DLLIMPORT_DECL_ATTRIBUTES 8817If the only target-specific handling you require is @samp{dllimport} 8818for Microsoft Windows targets, you should define the macro 8819@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler 8820will then define a function called 8821@code{merge_dllimport_decl_attributes} which can then be defined as 8822the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also 8823add @code{handle_dll_attribute} in the attribute table for your port 8824to perform initial processing of the @samp{dllimport} and 8825@samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and 8826@file{i386/i386.c}, for example. 8827@end deftypefn 8828 8829@deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (tree @var{decl}) 8830@var{decl} is a variable or function with @code{__attribute__((dllimport))} 8831specified. Use this hook if the target needs to add extra validation 8832checks to @code{handle_dll_attribute}. 8833@end deftypefn 8834 8835@defmac TARGET_DECLSPEC 8836Define this macro to a nonzero value if you want to treat 8837@code{__declspec(X)} as equivalent to @code{__attribute((X))}. By 8838default, this behavior is enabled only for targets that define 8839@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation 8840of @code{__declspec} is via a built-in macro, but you should not rely 8841on this implementation detail. 8842@end defmac 8843 8844@deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr}) 8845Define this target hook if you want to be able to add attributes to a decl 8846when it is being created. This is normally useful for back ends which 8847wish to implement a pragma by using the attributes which correspond to 8848the pragma's effect. The @var{node} argument is the decl which is being 8849created. The @var{attr_ptr} argument is a pointer to the attribute list 8850for this decl. The list itself should not be modified, since it may be 8851shared with other decls, but attributes may be chained on the head of 8852the list and @code{*@var{attr_ptr}} modified to point to the new 8853attributes, or a copy of the list may be made if further changes are 8854needed. 8855@end deftypefn 8856 8857@deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (tree @var{fndecl}) 8858@cindex inlining 8859This target hook returns @code{true} if it is ok to inline @var{fndecl} 8860into the current function, despite its having target-specific 8861attributes, @code{false} otherwise. By default, if a function has a 8862target specific attribute attached to it, it will not be inlined. 8863@end deftypefn 8864 8865@node MIPS Coprocessors 8866@section Defining coprocessor specifics for MIPS targets. 8867@cindex MIPS coprocessor-definition macros 8868 8869The MIPS specification allows MIPS implementations to have as many as 4 8870coprocessors, each with as many as 32 private registers. GCC supports 8871accessing these registers and transferring values between the registers 8872and memory using asm-ized variables. For example: 8873 8874@smallexample 8875 register unsigned int cp0count asm ("c0r1"); 8876 unsigned int d; 8877 8878 d = cp0count + 3; 8879@end smallexample 8880 8881(``c0r1'' is the default name of register 1 in coprocessor 0; alternate 8882names may be added as described below, or the default names may be 8883overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.) 8884 8885Coprocessor registers are assumed to be epilogue-used; sets to them will 8886be preserved even if it does not appear that the register is used again 8887later in the function. 8888 8889Another note: according to the MIPS spec, coprocessor 1 (if present) is 8890the FPU@. One accesses COP1 registers through standard mips 8891floating-point support; they are not included in this mechanism. 8892 8893There is one macro used in defining the MIPS coprocessor interface which 8894you may want to override in subtargets; it is described below. 8895 8896@defmac ALL_COP_ADDITIONAL_REGISTER_NAMES 8897A comma-separated list (with leading comma) of pairs describing the 8898alternate names of coprocessor registers. The format of each entry should be 8899@smallexample 8900@{ @var{alternatename}, @var{register_number}@} 8901@end smallexample 8902Default: empty. 8903@end defmac 8904 8905@node PCH Target 8906@section Parameters for Precompiled Header Validity Checking 8907@cindex parameters, precompiled headers 8908 8909@deftypefn {Target Hook} void *TARGET_GET_PCH_VALIDITY (size_t *@var{sz}) 8910This hook returns the data needed by @code{TARGET_PCH_VALID_P} and sets 8911@samp{*@var{sz}} to the size of the data in bytes. 8912@end deftypefn 8913 8914@deftypefn {Target Hook} const char *TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz}) 8915This hook checks whether the options used to create a PCH file are 8916compatible with the current settings. It returns @code{NULL} 8917if so and a suitable error message if not. Error messages will 8918be presented to the user and must be localized using @samp{_(@var{msg})}. 8919 8920@var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY} 8921when the PCH file was created and @var{sz} is the size of that data in bytes. 8922It's safe to assume that the data was created by the same version of the 8923compiler, so no format checking is needed. 8924 8925The default definition of @code{default_pch_valid_p} should be 8926suitable for most targets. 8927@end deftypefn 8928 8929@deftypefn {Target Hook} const char *TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags}) 8930If this hook is nonnull, the default implementation of 8931@code{TARGET_PCH_VALID_P} will use it to check for compatible values 8932of @code{target_flags}. @var{pch_flags} specifies the value that 8933@code{target_flags} had when the PCH file was created. The return 8934value is the same as for @code{TARGET_PCH_VALID_P}. 8935@end deftypefn 8936 8937@node C++ ABI 8938@section C++ ABI parameters 8939@cindex parameters, c++ abi 8940 8941@deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void) 8942Define this hook to override the integer type used for guard variables. 8943These are used to implement one-time construction of static objects. The 8944default is long_long_integer_type_node. 8945@end deftypefn 8946 8947@deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void) 8948This hook determines how guard variables are used. It should return 8949@code{false} (the default) if first byte should be used. A return value of 8950@code{true} indicates the least significant bit should be used. 8951@end deftypefn 8952 8953@deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type}) 8954This hook returns the size of the cookie to use when allocating an array 8955whose elements have the indicated @var{type}. Assumes that it is already 8956known that a cookie is needed. The default is 8957@code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the 8958IA64/Generic C++ ABI@. 8959@end deftypefn 8960 8961@deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void) 8962This hook should return @code{true} if the element size should be stored in 8963array cookies. The default is to return @code{false}. 8964@end deftypefn 8965 8966@deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export}) 8967If defined by a backend this hook allows the decision made to export 8968class @var{type} to be overruled. Upon entry @var{import_export} 8969will contain 1 if the class is going to be exported, @minus{}1 if it is going 8970to be imported and 0 otherwise. This function should return the 8971modified value and perform any other actions necessary to support the 8972backend's targeted operating system. 8973@end deftypefn 8974 8975@deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void) 8976This hook should return @code{true} if constructors and destructors return 8977the address of the object created/destroyed. The default is to return 8978@code{false}. 8979@end deftypefn 8980 8981@deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void) 8982This hook returns true if the key method for a class (i.e., the method 8983which, if defined in the current translation unit, causes the virtual 8984table to be emitted) may be an inline function. Under the standard 8985Itanium C++ ABI the key method may be an inline function so long as 8986the function is not declared inline in the class definition. Under 8987some variants of the ABI, an inline function can never be the key 8988method. The default is to return @code{true}. 8989@end deftypefn 8990 8991@deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl}) 8992@var{decl} is a virtual table, virtual table table, typeinfo object, 8993or other similar implicit class data object that will be emitted with 8994external linkage in this translation unit. No ELF visibility has been 8995explicitly specified. If the target needs to specify a visibility 8996other than that of the containing class, use this hook to set 8997@code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}. 8998@end deftypefn 8999 9000@deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void) 9001This hook returns true (the default) if virtual tables and other 9002similar implicit class data objects are always COMDAT if they have 9003external linkage. If this hook returns false, then class data for 9004classes whose virtual table will be emitted in only one translation 9005unit will not be COMDAT. 9006@end deftypefn 9007 9008@deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void) 9009This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI) 9010should be used to register static destructors when @option{-fuse-cxa-atexit} 9011is in effect. The default is to return false to use @code{__cxa_atexit}. 9012@end deftypefn 9013 9014@deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type}) 9015@var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just been 9016defined. Use this hook to make adjustments to the class (eg, tweak 9017visibility or perform any other required target modifications). 9018@end deftypefn 9019 9020@node Misc 9021@section Miscellaneous Parameters 9022@cindex parameters, miscellaneous 9023 9024@c prevent bad page break with this line 9025Here are several miscellaneous parameters. 9026 9027@defmac HAS_LONG_COND_BRANCH 9028Define this boolean macro to indicate whether or not your architecture 9029has conditional branches that can span all of memory. It is used in 9030conjunction with an optimization that partitions hot and cold basic 9031blocks into separate sections of the executable. If this macro is 9032set to false, gcc will convert any conditional branches that attempt 9033to cross between sections into unconditional branches or indirect jumps. 9034@end defmac 9035 9036@defmac HAS_LONG_UNCOND_BRANCH 9037Define this boolean macro to indicate whether or not your architecture 9038has unconditional branches that can span all of memory. It is used in 9039conjunction with an optimization that partitions hot and cold basic 9040blocks into separate sections of the executable. If this macro is 9041set to false, gcc will convert any unconditional branches that attempt 9042to cross between sections into indirect jumps. 9043@end defmac 9044 9045@defmac CASE_VECTOR_MODE 9046An alias for a machine mode name. This is the machine mode that 9047elements of a jump-table should have. 9048@end defmac 9049 9050@defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body}) 9051Optional: return the preferred mode for an @code{addr_diff_vec} 9052when the minimum and maximum offset are known. If you define this, 9053it enables extra code in branch shortening to deal with @code{addr_diff_vec}. 9054To make this work, you also have to define @code{INSN_ALIGN} and 9055make the alignment for @code{addr_diff_vec} explicit. 9056The @var{body} argument is provided so that the offset_unsigned and scale 9057flags can be updated. 9058@end defmac 9059 9060@defmac CASE_VECTOR_PC_RELATIVE 9061Define this macro to be a C expression to indicate when jump-tables 9062should contain relative addresses. You need not define this macro if 9063jump-tables never contain relative addresses, or jump-tables should 9064contain relative addresses only when @option{-fPIC} or @option{-fPIC} 9065is in effect. 9066@end defmac 9067 9068@defmac CASE_VALUES_THRESHOLD 9069Define this to be the smallest number of different values for which it 9070is best to use a jump-table instead of a tree of conditional branches. 9071The default is four for machines with a @code{casesi} instruction and 9072five otherwise. This is best for most machines. 9073@end defmac 9074 9075@defmac CASE_USE_BIT_TESTS 9076Define this macro to be a C expression to indicate whether C switch 9077statements may be implemented by a sequence of bit tests. This is 9078advantageous on processors that can efficiently implement left shift 9079of 1 by the number of bits held in a register, but inappropriate on 9080targets that would require a loop. By default, this macro returns 9081@code{true} if the target defines an @code{ashlsi3} pattern, and 9082@code{false} otherwise. 9083@end defmac 9084 9085@defmac WORD_REGISTER_OPERATIONS 9086Define this macro if operations between registers with integral mode 9087smaller than a word are always performed on the entire register. 9088Most RISC machines have this property and most CISC machines do not. 9089@end defmac 9090 9091@defmac LOAD_EXTEND_OP (@var{mem_mode}) 9092Define this macro to be a C expression indicating when insns that read 9093memory in @var{mem_mode}, an integral mode narrower than a word, set the 9094bits outside of @var{mem_mode} to be either the sign-extension or the 9095zero-extension of the data read. Return @code{SIGN_EXTEND} for values 9096of @var{mem_mode} for which the 9097insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and 9098@code{UNKNOWN} for other modes. 9099 9100This macro is not called with @var{mem_mode} non-integral or with a width 9101greater than or equal to @code{BITS_PER_WORD}, so you may return any 9102value in this case. Do not define this macro if it would always return 9103@code{UNKNOWN}. On machines where this macro is defined, you will normally 9104define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}. 9105 9106You may return a non-@code{UNKNOWN} value even if for some hard registers 9107the sign extension is not performed, if for the @code{REGNO_REG_CLASS} 9108of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero 9109when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any 9110integral mode larger than this but not larger than @code{word_mode}. 9111 9112You must return @code{UNKNOWN} if for some hard registers that allow this 9113mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to 9114@code{word_mode}, but that they can change to another integral mode that 9115is larger then @var{mem_mode} but still smaller than @code{word_mode}. 9116@end defmac 9117 9118@defmac SHORT_IMMEDIATES_SIGN_EXTEND 9119Define this macro if loading short immediate values into registers sign 9120extends. 9121@end defmac 9122 9123@defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC 9124Define this macro if the same instructions that convert a floating 9125point number to a signed fixed point number also convert validly to an 9126unsigned one. 9127@end defmac 9128 9129@deftypefn {Target Hook} int TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (enum machine_mode @var{mode}) 9130When @option{-ffast-math} is in effect, GCC tries to optimize 9131divisions by the same divisor, by turning them into multiplications by 9132the reciprocal. This target hook specifies the minimum number of divisions 9133that should be there for GCC to perform the optimization for a variable 9134of mode @var{mode}. The default implementation returns 3 if the machine 9135has an instruction for the division, and 2 if it does not. 9136@end deftypefn 9137 9138@defmac MOVE_MAX 9139The maximum number of bytes that a single instruction can move quickly 9140between memory and registers or between two memory locations. 9141@end defmac 9142 9143@defmac MAX_MOVE_MAX 9144The maximum number of bytes that a single instruction can move quickly 9145between memory and registers or between two memory locations. If this 9146is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the 9147constant value that is the largest value that @code{MOVE_MAX} can have 9148at run-time. 9149@end defmac 9150 9151@defmac SHIFT_COUNT_TRUNCATED 9152A C expression that is nonzero if on this machine the number of bits 9153actually used for the count of a shift operation is equal to the number 9154of bits needed to represent the size of the object being shifted. When 9155this macro is nonzero, the compiler will assume that it is safe to omit 9156a sign-extend, zero-extend, and certain bitwise `and' instructions that 9157truncates the count of a shift operation. On machines that have 9158instructions that act on bit-fields at variable positions, which may 9159include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED} 9160also enables deletion of truncations of the values that serve as 9161arguments to bit-field instructions. 9162 9163If both types of instructions truncate the count (for shifts) and 9164position (for bit-field operations), or if no variable-position bit-field 9165instructions exist, you should define this macro. 9166 9167However, on some machines, such as the 80386 and the 680x0, truncation 9168only applies to shift operations and not the (real or pretended) 9169bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on 9170such machines. Instead, add patterns to the @file{md} file that include 9171the implied truncation of the shift instructions. 9172 9173You need not define this macro if it would always have the value of zero. 9174@end defmac 9175 9176@anchor{TARGET_SHIFT_TRUNCATION_MASK} 9177@deftypefn {Target Hook} int TARGET_SHIFT_TRUNCATION_MASK (enum machine_mode @var{mode}) 9178This function describes how the standard shift patterns for @var{mode} 9179deal with shifts by negative amounts or by more than the width of the mode. 9180@xref{shift patterns}. 9181 9182On many machines, the shift patterns will apply a mask @var{m} to the 9183shift count, meaning that a fixed-width shift of @var{x} by @var{y} is 9184equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If 9185this is true for mode @var{mode}, the function should return @var{m}, 9186otherwise it should return 0. A return value of 0 indicates that no 9187particular behavior is guaranteed. 9188 9189Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does 9190@emph{not} apply to general shift rtxes; it applies only to instructions 9191that are generated by the named shift patterns. 9192 9193The default implementation of this function returns 9194@code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED} 9195and 0 otherwise. This definition is always safe, but if 9196@code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns 9197nevertheless truncate the shift count, you may get better code 9198by overriding it. 9199@end deftypefn 9200 9201@defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec}) 9202A C expression which is nonzero if on this machine it is safe to 9203``convert'' an integer of @var{inprec} bits to one of @var{outprec} 9204bits (where @var{outprec} is smaller than @var{inprec}) by merely 9205operating on it as if it had only @var{outprec} bits. 9206 9207On many machines, this expression can be 1. 9208 9209@c rearranged this, removed the phrase "it is reported that". this was 9210@c to fix an overfull hbox. --mew 10feb93 9211When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for 9212modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result. 9213If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in 9214such cases may improve things. 9215@end defmac 9216 9217@deftypefn {Target Hook} int TARGET_MODE_REP_EXTENDED (enum machine_mode @var{mode}, enum machine_mode @var{rep_mode}) 9218The representation of an integral mode can be such that the values 9219are always extended to a wider integral mode. Return 9220@code{SIGN_EXTEND} if values of @var{mode} are represented in 9221sign-extended form to @var{rep_mode}. Return @code{UNKNOWN} 9222otherwise. (Currently, none of the targets use zero-extended 9223representation this way so unlike @code{LOAD_EXTEND_OP}, 9224@code{TARGET_MODE_REP_EXTENDED} is expected to return either 9225@code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends 9226@var{mode} to @var{mode_rep} so that @var{mode_rep} is not the next 9227widest integral mode and currently we take advantage of this fact.) 9228 9229Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN} 9230value even if the extension is not performed on certain hard registers 9231as long as for the @code{REGNO_REG_CLASS} of these hard registers 9232@code{CANNOT_CHANGE_MODE_CLASS} returns nonzero. 9233 9234Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP} 9235describe two related properties. If you define 9236@code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want 9237to define @code{LOAD_EXTEND_OP (mode)} to return the same type of 9238extension. 9239 9240In order to enforce the representation of @code{mode}, 9241@code{TRULY_NOOP_TRUNCATION} should return false when truncating to 9242@code{mode}. 9243@end deftypefn 9244 9245@defmac STORE_FLAG_VALUE 9246A C expression describing the value returned by a comparison operator 9247with an integral mode and stored by a store-flag instruction 9248(@samp{s@var{cond}}) when the condition is true. This description must 9249apply to @emph{all} the @samp{s@var{cond}} patterns and all the 9250comparison operators whose results have a @code{MODE_INT} mode. 9251 9252A value of 1 or @minus{}1 means that the instruction implementing the 9253comparison operator returns exactly 1 or @minus{}1 when the comparison is true 9254and 0 when the comparison is false. Otherwise, the value indicates 9255which bits of the result are guaranteed to be 1 when the comparison is 9256true. This value is interpreted in the mode of the comparison 9257operation, which is given by the mode of the first operand in the 9258@samp{s@var{cond}} pattern. Either the low bit or the sign bit of 9259@code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by 9260the compiler. 9261 9262If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will 9263generate code that depends only on the specified bits. It can also 9264replace comparison operators with equivalent operations if they cause 9265the required bits to be set, even if the remaining bits are undefined. 9266For example, on a machine whose comparison operators return an 9267@code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as 9268@samp{0x80000000}, saying that just the sign bit is relevant, the 9269expression 9270 9271@smallexample 9272(ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0)) 9273@end smallexample 9274 9275@noindent 9276can be converted to 9277 9278@smallexample 9279(ashift:SI @var{x} (const_int @var{n})) 9280@end smallexample 9281 9282@noindent 9283where @var{n} is the appropriate shift count to move the bit being 9284tested into the sign bit. 9285 9286There is no way to describe a machine that always sets the low-order bit 9287for a true value, but does not guarantee the value of any other bits, 9288but we do not know of any machine that has such an instruction. If you 9289are trying to port GCC to such a machine, include an instruction to 9290perform a logical-and of the result with 1 in the pattern for the 9291comparison operators and let us know at @email{gcc@@gcc.gnu.org}. 9292 9293Often, a machine will have multiple instructions that obtain a value 9294from a comparison (or the condition codes). Here are rules to guide the 9295choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions 9296to be used: 9297 9298@itemize @bullet 9299@item 9300Use the shortest sequence that yields a valid definition for 9301@code{STORE_FLAG_VALUE}. It is more efficient for the compiler to 9302``normalize'' the value (convert it to, e.g., 1 or 0) than for the 9303comparison operators to do so because there may be opportunities to 9304combine the normalization with other operations. 9305 9306@item 9307For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being 9308slightly preferred on machines with expensive jumps and 1 preferred on 9309other machines. 9310 9311@item 9312As a second choice, choose a value of @samp{0x80000001} if instructions 9313exist that set both the sign and low-order bits but do not define the 9314others. 9315 9316@item 9317Otherwise, use a value of @samp{0x80000000}. 9318@end itemize 9319 9320Many machines can produce both the value chosen for 9321@code{STORE_FLAG_VALUE} and its negation in the same number of 9322instructions. On those machines, you should also define a pattern for 9323those cases, e.g., one matching 9324 9325@smallexample 9326(set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C}))) 9327@end smallexample 9328 9329Some machines can also perform @code{and} or @code{plus} operations on 9330condition code values with less instructions than the corresponding 9331@samp{s@var{cond}} insn followed by @code{and} or @code{plus}. On those 9332machines, define the appropriate patterns. Use the names @code{incscc} 9333and @code{decscc}, respectively, for the patterns which perform 9334@code{plus} or @code{minus} operations on condition code values. See 9335@file{rs6000.md} for some examples. The GNU Superoptizer can be used to 9336find such instruction sequences on other machines. 9337 9338If this macro is not defined, the default value, 1, is used. You need 9339not define @code{STORE_FLAG_VALUE} if the machine has no store-flag 9340instructions, or if the value generated by these instructions is 1. 9341@end defmac 9342 9343@defmac FLOAT_STORE_FLAG_VALUE (@var{mode}) 9344A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is 9345returned when comparison operators with floating-point results are true. 9346Define this macro on machines that have comparison operations that return 9347floating-point values. If there are no such operations, do not define 9348this macro. 9349@end defmac 9350 9351@defmac VECTOR_STORE_FLAG_VALUE (@var{mode}) 9352A C expression that gives a rtx representing the nonzero true element 9353for vector comparisons. The returned rtx should be valid for the inner 9354mode of @var{mode} which is guaranteed to be a vector mode. Define 9355this macro on machines that have vector comparison operations that 9356return a vector result. If there are no such operations, do not define 9357this macro. Typically, this macro is defined as @code{const1_rtx} or 9358@code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent 9359the compiler optimizing such vector comparison operations for the 9360given mode. 9361@end defmac 9362 9363@defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 9364@defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 9365A C expression that evaluates to true if the architecture defines a value 9366for @code{clz} or @code{ctz} with a zero operand. If so, @var{value} 9367should be set to this value. If this macro is not defined, the value of 9368@code{clz} or @code{ctz} is assumed to be undefined. 9369 9370This macro must be defined if the target's expansion for @code{ffs} 9371relies on a particular value to get correct results. Otherwise it 9372is not necessary, though it may be used to optimize some corner cases. 9373 9374Note that regardless of this macro the ``definedness'' of @code{clz} 9375and @code{ctz} at zero do @emph{not} extend to the builtin functions 9376visible to the user. Thus one may be free to adjust the value at will 9377to match the target expansion of these operations without fear of 9378breaking the API@. 9379@end defmac 9380 9381@defmac Pmode 9382An alias for the machine mode for pointers. On most machines, define 9383this to be the integer mode corresponding to the width of a hardware 9384pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines. 9385On some machines you must define this to be one of the partial integer 9386modes, such as @code{PSImode}. 9387 9388The width of @code{Pmode} must be at least as large as the value of 9389@code{POINTER_SIZE}. If it is not equal, you must define the macro 9390@code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended 9391to @code{Pmode}. 9392@end defmac 9393 9394@defmac FUNCTION_MODE 9395An alias for the machine mode used for memory references to functions 9396being called, in @code{call} RTL expressions. On most machines this 9397should be @code{QImode}. 9398@end defmac 9399 9400@defmac STDC_0_IN_SYSTEM_HEADERS 9401In normal operation, the preprocessor expands @code{__STDC__} to the 9402constant 1, to signify that GCC conforms to ISO Standard C@. On some 9403hosts, like Solaris, the system compiler uses a different convention, 9404where @code{__STDC__} is normally 0, but is 1 if the user specifies 9405strict conformance to the C Standard. 9406 9407Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host 9408convention when processing system header files, but when processing user 9409files @code{__STDC__} will always expand to 1. 9410@end defmac 9411 9412@defmac NO_IMPLICIT_EXTERN_C 9413Define this macro if the system header files support C++ as well as C@. 9414This macro inhibits the usual method of using system header files in 9415C++, which is to pretend that the file's contents are enclosed in 9416@samp{extern "C" @{@dots{}@}}. 9417@end defmac 9418 9419@findex #pragma 9420@findex pragma 9421@defmac REGISTER_TARGET_PRAGMAS () 9422Define this macro if you want to implement any target-specific pragmas. 9423If defined, it is a C expression which makes a series of calls to 9424@code{c_register_pragma} or @code{c_register_pragma_with_expansion} 9425for each pragma. The macro may also do any 9426setup required for the pragmas. 9427 9428The primary reason to define this macro is to provide compatibility with 9429other compilers for the same target. In general, we discourage 9430definition of target-specific pragmas for GCC@. 9431 9432If the pragma can be implemented by attributes then you should consider 9433defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well. 9434 9435Preprocessor macros that appear on pragma lines are not expanded. All 9436@samp{#pragma} directives that do not match any registered pragma are 9437silently ignored, unless the user specifies @option{-Wunknown-pragmas}. 9438@end defmac 9439 9440@deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 9441@deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 9442 9443Each call to @code{c_register_pragma} or 9444@code{c_register_pragma_with_expansion} establishes one pragma. The 9445@var{callback} routine will be called when the preprocessor encounters a 9446pragma of the form 9447 9448@smallexample 9449#pragma [@var{space}] @var{name} @dots{} 9450@end smallexample 9451 9452@var{space} is the case-sensitive namespace of the pragma, or 9453@code{NULL} to put the pragma in the global namespace. The callback 9454routine receives @var{pfile} as its first argument, which can be passed 9455on to cpplib's functions if necessary. You can lex tokens after the 9456@var{name} by calling @code{pragma_lex}. Tokens that are not read by the 9457callback will be silently ignored. The end of the line is indicated by 9458a token of type @code{CPP_EOF}. Macro expansion occurs on the 9459arguments of pragmas registered with 9460@code{c_register_pragma_with_expansion} but not on the arguments of 9461pragmas registered with @code{c_register_pragma}. 9462 9463For an example use of this routine, see @file{c4x.h} and the callback 9464routines defined in @file{c4x-c.c}. 9465 9466Note that the use of @code{pragma_lex} is specific to the C and C++ 9467compilers. It will not work in the Java or Fortran compilers, or any 9468other language compilers for that matter. Thus if @code{pragma_lex} is going 9469to be called from target-specific code, it must only be done so when 9470building the C and C++ compilers. This can be done by defining the 9471variables @code{c_target_objs} and @code{cxx_target_objs} in the 9472target entry in the @file{config.gcc} file. These variables should name 9473the target-specific, language-specific object file which contains the 9474code that uses @code{pragma_lex}. Note it will also be necessary to add a 9475rule to the makefile fragment pointed to by @code{tmake_file} that shows 9476how to build this object file. 9477@end deftypefun 9478 9479@findex #pragma 9480@findex pragma 9481@defmac HANDLE_SYSV_PRAGMA 9482Define this macro (to a value of 1) if you want the System V style 9483pragmas @samp{#pragma pack(<n>)} and @samp{#pragma weak <name> 9484[=<value>]} to be supported by gcc. 9485 9486The pack pragma specifies the maximum alignment (in bytes) of fields 9487within a structure, in much the same way as the @samp{__aligned__} and 9488@samp{__packed__} @code{__attribute__}s do. A pack value of zero resets 9489the behavior to the default. 9490 9491A subtlety for Microsoft Visual C/C++ style bit-field packing 9492(e.g.@: -mms-bitfields) for targets that support it: 9493When a bit-field is inserted into a packed record, the whole size 9494of the underlying type is used by one or more same-size adjacent 9495bit-fields (that is, if its long:3, 32 bits is used in the record, 9496and any additional adjacent long bit-fields are packed into the same 9497chunk of 32 bits. However, if the size changes, a new field of that 9498size is allocated). 9499 9500If both MS bit-fields and @samp{__attribute__((packed))} are used, 9501the latter will take precedence. If @samp{__attribute__((packed))} is 9502used on a single field when MS bit-fields are in use, it will take 9503precedence for that field, but the alignment of the rest of the structure 9504may affect its placement. 9505 9506The weak pragma only works if @code{SUPPORTS_WEAK} and 9507@code{ASM_WEAKEN_LABEL} are defined. If enabled it allows the creation 9508of specifically named weak labels, optionally with a value. 9509@end defmac 9510 9511@findex #pragma 9512@findex pragma 9513@defmac HANDLE_PRAGMA_PACK_PUSH_POP 9514Define this macro (to a value of 1) if you want to support the Win32 9515style pragmas @samp{#pragma pack(push[,@var{n}])} and @samp{#pragma 9516pack(pop)}. The @samp{pack(push,[@var{n}])} pragma specifies the maximum 9517alignment (in bytes) of fields within a structure, in much the same way as 9518the @samp{__aligned__} and @samp{__packed__} @code{__attribute__}s do. A 9519pack value of zero resets the behavior to the default. Successive 9520invocations of this pragma cause the previous values to be stacked, so 9521that invocations of @samp{#pragma pack(pop)} will return to the previous 9522value. 9523@end defmac 9524 9525@defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION 9526Define this macro, as well as 9527@code{HANDLE_SYSV_PRAGMA}, if macros should be expanded in the 9528arguments of @samp{#pragma pack}. 9529@end defmac 9530 9531@defmac TARGET_DEFAULT_PACK_STRUCT 9532If your target requires a structure packing default other than 0 (meaning 9533the machine default), define this macro to the necessary value (in bytes). 9534This must be a value that would also be valid to use with 9535@samp{#pragma pack()} (that is, a small power of two). 9536@end defmac 9537 9538@defmac DOLLARS_IN_IDENTIFIERS 9539Define this macro to control use of the character @samp{$} in 9540identifier names for the C family of languages. 0 means @samp{$} is 9541not allowed by default; 1 means it is allowed. 1 is the default; 9542there is no need to define this macro in that case. 9543@end defmac 9544 9545@defmac NO_DOLLAR_IN_LABEL 9546Define this macro if the assembler does not accept the character 9547@samp{$} in label names. By default constructors and destructors in 9548G++ have @samp{$} in the identifiers. If this macro is defined, 9549@samp{.} is used instead. 9550@end defmac 9551 9552@defmac NO_DOT_IN_LABEL 9553Define this macro if the assembler does not accept the character 9554@samp{.} in label names. By default constructors and destructors in G++ 9555have names that use @samp{.}. If this macro is defined, these names 9556are rewritten to avoid @samp{.}. 9557@end defmac 9558 9559@defmac INSN_SETS_ARE_DELAYED (@var{insn}) 9560Define this macro as a C expression that is nonzero if it is safe for the 9561delay slot scheduler to place instructions in the delay slot of @var{insn}, 9562even if they appear to use a resource set or clobbered in @var{insn}. 9563@var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that 9564every @code{call_insn} has this behavior. On machines where some @code{insn} 9565or @code{jump_insn} is really a function call and hence has this behavior, 9566you should define this macro. 9567 9568You need not define this macro if it would always return zero. 9569@end defmac 9570 9571@defmac INSN_REFERENCES_ARE_DELAYED (@var{insn}) 9572Define this macro as a C expression that is nonzero if it is safe for the 9573delay slot scheduler to place instructions in the delay slot of @var{insn}, 9574even if they appear to set or clobber a resource referenced in @var{insn}. 9575@var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where 9576some @code{insn} or @code{jump_insn} is really a function call and its operands 9577are registers whose use is actually in the subroutine it calls, you should 9578define this macro. Doing so allows the delay slot scheduler to move 9579instructions which copy arguments into the argument registers into the delay 9580slot of @var{insn}. 9581 9582You need not define this macro if it would always return zero. 9583@end defmac 9584 9585@defmac MULTIPLE_SYMBOL_SPACES 9586Define this macro as a C expression that is nonzero if, in some cases, 9587global symbols from one translation unit may not be bound to undefined 9588symbols in another translation unit without user intervention. For 9589instance, under Microsoft Windows symbols must be explicitly imported 9590from shared libraries (DLLs). 9591 9592You need not define this macro if it would always evaluate to zero. 9593@end defmac 9594 9595@deftypefn {Target Hook} tree TARGET_MD_ASM_CLOBBERS (tree @var{outputs}, tree @var{inputs}, tree @var{clobbers}) 9596This target hook should add to @var{clobbers} @code{STRING_CST} trees for 9597any hard regs the port wishes to automatically clobber for an asm. 9598It should return the result of the last @code{tree_cons} used to add a 9599clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the 9600corresponding parameters to the asm and may be inspected to avoid 9601clobbering a register that is an input or output of the asm. You can use 9602@code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test 9603for overlap with regards to asm-declared registers. 9604@end deftypefn 9605 9606@defmac MATH_LIBRARY 9607Define this macro as a C string constant for the linker argument to link 9608in the system math library, or @samp{""} if the target does not have a 9609separate math library. 9610 9611You need only define this macro if the default of @samp{"-lm"} is wrong. 9612@end defmac 9613 9614@defmac LIBRARY_PATH_ENV 9615Define this macro as a C string constant for the environment variable that 9616specifies where the linker should look for libraries. 9617 9618You need only define this macro if the default of @samp{"LIBRARY_PATH"} 9619is wrong. 9620@end defmac 9621 9622@defmac TARGET_POSIX_IO 9623Define this macro if the target supports the following POSIX@ file 9624functions, access, mkdir and file locking with fcntl / F_SETLKW@. 9625Defining @code{TARGET_POSIX_IO} will enable the test coverage code 9626to use file locking when exiting a program, which avoids race conditions 9627if the program has forked. It will also create directories at run-time 9628for cross-profiling. 9629@end defmac 9630 9631@defmac MAX_CONDITIONAL_EXECUTE 9632 9633A C expression for the maximum number of instructions to execute via 9634conditional execution instructions instead of a branch. A value of 9635@code{BRANCH_COST}+1 is the default if the machine does not use cc0, and 96361 if it does use cc0. 9637@end defmac 9638 9639@defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr}) 9640Used if the target needs to perform machine-dependent modifications on the 9641conditionals used for turning basic blocks into conditionally executed code. 9642@var{ce_info} points to a data structure, @code{struct ce_if_block}, which 9643contains information about the currently processed blocks. @var{true_expr} 9644and @var{false_expr} are the tests that are used for converting the 9645then-block and the else-block, respectively. Set either @var{true_expr} or 9646@var{false_expr} to a null pointer if the tests cannot be converted. 9647@end defmac 9648 9649@defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr}) 9650Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated 9651if-statements into conditions combined by @code{and} and @code{or} operations. 9652@var{bb} contains the basic block that contains the test that is currently 9653being processed and about to be turned into a condition. 9654@end defmac 9655 9656@defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn}) 9657A C expression to modify the @var{PATTERN} of an @var{INSN} that is to 9658be converted to conditional execution format. @var{ce_info} points to 9659a data structure, @code{struct ce_if_block}, which contains information 9660about the currently processed blocks. 9661@end defmac 9662 9663@defmac IFCVT_MODIFY_FINAL (@var{ce_info}) 9664A C expression to perform any final machine dependent modifications in 9665converting code to conditional execution. The involved basic blocks 9666can be found in the @code{struct ce_if_block} structure that is pointed 9667to by @var{ce_info}. 9668@end defmac 9669 9670@defmac IFCVT_MODIFY_CANCEL (@var{ce_info}) 9671A C expression to cancel any machine dependent modifications in 9672converting code to conditional execution. The involved basic blocks 9673can be found in the @code{struct ce_if_block} structure that is pointed 9674to by @var{ce_info}. 9675@end defmac 9676 9677@defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info}) 9678A C expression to initialize any extra fields in a @code{struct ce_if_block} 9679structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro. 9680@end defmac 9681 9682@defmac IFCVT_EXTRA_FIELDS 9683If defined, it should expand to a set of field declarations that will be 9684added to the @code{struct ce_if_block} structure. These should be initialized 9685by the @code{IFCVT_INIT_EXTRA_FIELDS} macro. 9686@end defmac 9687 9688@deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG () 9689If non-null, this hook performs a target-specific pass over the 9690instruction stream. The compiler will run it at all optimization levels, 9691just before the point at which it normally does delayed-branch scheduling. 9692 9693The exact purpose of the hook varies from target to target. Some use 9694it to do transformations that are necessary for correctness, such as 9695laying out in-function constant pools or avoiding hardware hazards. 9696Others use it as an opportunity to do some machine-dependent optimizations. 9697 9698You need not implement the hook if it has nothing to do. The default 9699definition is null. 9700@end deftypefn 9701 9702@deftypefn {Target Hook} void TARGET_INIT_BUILTINS () 9703Define this hook if you have any machine-specific built-in functions 9704that need to be defined. It should be a function that performs the 9705necessary setup. 9706 9707Machine specific built-in functions can be useful to expand special machine 9708instructions that would otherwise not normally be generated because 9709they have no equivalent in the source language (for example, SIMD vector 9710instructions or prefetch instructions). 9711 9712To create a built-in function, call the function 9713@code{lang_hooks.builtin_function} 9714which is defined by the language front end. You can use any type nodes set 9715up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2}; 9716only language front ends that use those two functions will call 9717@samp{TARGET_INIT_BUILTINS}. 9718@end deftypefn 9719 9720@deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, enum machine_mode @var{mode}, int @var{ignore}) 9721 9722Expand a call to a machine specific built-in function that was set up by 9723@samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the 9724function call; the result should go to @var{target} if that is 9725convenient, and have mode @var{mode} if that is convenient. 9726@var{subtarget} may be used as the target for computing one of 9727@var{exp}'s operands. @var{ignore} is nonzero if the value is to be 9728ignored. This function should return the result of the call to the 9729built-in function. 9730@end deftypefn 9731 9732@deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (tree @var{fndecl}, tree @var{arglist}) 9733 9734Select a replacement for a machine specific built-in function that 9735was set up by @samp{TARGET_INIT_BUILTINS}. This is done 9736@emph{before} regular type checking, and so allows the target to 9737implement a crude form of function overloading. @var{fndecl} is the 9738declaration of the built-in function. @var{arglist} is the list of 9739arguments passed to the built-in function. The result is a 9740complete expression that implements the operation, usually 9741another @code{CALL_EXPR}. 9742@end deftypefn 9743 9744@deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, tree @var{arglist}, bool @var{ignore}) 9745 9746Fold a call to a machine specific built-in function that was set up by 9747@samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the 9748built-in function. @var{arglist} is the list of arguments passed to 9749the built-in function. The result is another tree containing a 9750simplified expression for the call's result. If @var{ignore} is true 9751the value will be ignored. 9752@end deftypefn 9753 9754@deftypefn {Target Hook} const char * TARGET_INVALID_WITHIN_DOLOOP (rtx @var{insn}) 9755 9756Take an instruction in @var{insn} and return NULL if it is valid within a 9757low-overhead loop, otherwise return a string why doloop could not be applied. 9758 9759Many targets use special registers for low-overhead looping. For any 9760instruction that clobbers these this function should return a string indicating 9761the reason why the doloop could not be applied. 9762By default, the RTL loop optimizer does not use a present doloop pattern for 9763loops containing function calls or branch on table instructions. 9764@end deftypefn 9765 9766@defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2}) 9767 9768Take a branch insn in @var{branch1} and another in @var{branch2}. 9769Return true if redirecting @var{branch1} to the destination of 9770@var{branch2} is possible. 9771 9772On some targets, branches may have a limited range. Optimizing the 9773filling of delay slots can result in branches being redirected, and this 9774may in turn cause a branch offset to overflow. 9775@end defmac 9776 9777@deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (rtx @var{x}, @var{outer_code}) 9778This target hook returns @code{true} if @var{x} is considered to be commutative. 9779Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider 9780PLUS to be commutative inside a MEM. @var{outer_code} is the rtx code 9781of the enclosing rtl, if known, otherwise it is UNKNOWN. 9782@end deftypefn 9783 9784@deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg}) 9785 9786When the initial value of a hard register has been copied in a pseudo 9787register, it is often not necessary to actually allocate another register 9788to this pseudo register, because the original hard register or a stack slot 9789it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE} 9790is called at the start of register allocation once for each hard register 9791that had its initial value copied by using 9792@code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}. 9793Possible values are @code{NULL_RTX}, if you don't want 9794to do any special allocation, a @code{REG} rtx---that would typically be 9795the hard register itself, if it is known not to be clobbered---or a 9796@code{MEM}. 9797If you are returning a @code{MEM}, this is only a hint for the allocator; 9798it might decide to use another register anyways. 9799You may use @code{current_function_leaf_function} in the hook, functions 9800that use @code{REG_N_SETS}, to determine if the hard 9801register in question will not be clobbered. 9802The default value of this hook is @code{NULL}, which disables any special 9803allocation. 9804@end deftypefn 9805 9806@defmac TARGET_OBJECT_SUFFIX 9807Define this macro to be a C string representing the suffix for object 9808files on your target machine. If you do not define this macro, GCC will 9809use @samp{.o} as the suffix for object files. 9810@end defmac 9811 9812@defmac TARGET_EXECUTABLE_SUFFIX 9813Define this macro to be a C string representing the suffix to be 9814automatically added to executable files on your target machine. If you 9815do not define this macro, GCC will use the null string as the suffix for 9816executable files. 9817@end defmac 9818 9819@defmac COLLECT_EXPORT_LIST 9820If defined, @code{collect2} will scan the individual object files 9821specified on its command line and create an export list for the linker. 9822Define this macro for systems like AIX, where the linker discards 9823object files that are not referenced from @code{main} and uses export 9824lists. 9825@end defmac 9826 9827@defmac MODIFY_JNI_METHOD_CALL (@var{mdecl}) 9828Define this macro to a C expression representing a variant of the 9829method call @var{mdecl}, if Java Native Interface (JNI) methods 9830must be invoked differently from other methods on your target. 9831For example, on 32-bit Microsoft Windows, JNI methods must be invoked using 9832the @code{stdcall} calling convention and this macro is then 9833defined as this expression: 9834 9835@smallexample 9836build_type_attribute_variant (@var{mdecl}, 9837 build_tree_list 9838 (get_identifier ("stdcall"), 9839 NULL)) 9840@end smallexample 9841@end defmac 9842 9843@deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void) 9844This target hook returns @code{true} past the point in which new jump 9845instructions could be created. On machines that require a register for 9846every jump such as the SHmedia ISA of SH5, this point would typically be 9847reload, so this target hook should be defined to a function such as: 9848 9849@smallexample 9850static bool 9851cannot_modify_jumps_past_reload_p () 9852@{ 9853 return (reload_completed || reload_in_progress); 9854@} 9855@end smallexample 9856@end deftypefn 9857 9858@deftypefn {Target Hook} int TARGET_BRANCH_TARGET_REGISTER_CLASS (void) 9859This target hook returns a register class for which branch target register 9860optimizations should be applied. All registers in this class should be 9861usable interchangeably. After reload, registers in this class will be 9862re-allocated and loads will be hoisted out of loops and be subjected 9863to inter-block scheduling. 9864@end deftypefn 9865 9866@deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen}) 9867Branch target register optimization will by default exclude callee-saved 9868registers 9869that are not already live during the current function; if this target hook 9870returns true, they will be included. The target code must than make sure 9871that all target registers in the class returned by 9872@samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are 9873saved. @var{after_prologue_epilogue_gen} indicates if prologues and 9874epilogues have already been generated. Note, even if you only return 9875true when @var{after_prologue_epilogue_gen} is false, you still are likely 9876to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET} 9877to reserve space for caller-saved target registers. 9878@end deftypefn 9879 9880@defmac POWI_MAX_MULTS 9881If defined, this macro is interpreted as a signed integer C expression 9882that specifies the maximum number of floating point multiplications 9883that should be emitted when expanding exponentiation by an integer 9884constant inline. When this value is defined, exponentiation requiring 9885more than this number of multiplications is implemented by calling the 9886system library's @code{pow}, @code{powf} or @code{powl} routines. 9887The default value places no upper bound on the multiplication count. 9888@end defmac 9889 9890@deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc}) 9891This target hook should register any extra include files for the 9892target. The parameter @var{stdinc} indicates if normal include files 9893are present. The parameter @var{sysroot} is the system root directory. 9894The parameter @var{iprefix} is the prefix for the gcc directory. 9895@end deftypefn 9896 9897@deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc}) 9898This target hook should register any extra include files for the 9899target before any standard headers. The parameter @var{stdinc} 9900indicates if normal include files are present. The parameter 9901@var{sysroot} is the system root directory. The parameter 9902@var{iprefix} is the prefix for the gcc directory. 9903@end deftypefn 9904 9905@deftypefn Macro void TARGET_OPTF (char *@var{path}) 9906This target hook should register special include paths for the target. 9907The parameter @var{path} is the include to register. On Darwin 9908systems, this is used for Framework includes, which have semantics 9909that are different from @option{-I}. 9910@end deftypefn 9911 9912@deftypefn {Target Hook} bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl}) 9913This target hook returns @code{true} if it is safe to use a local alias 9914for a virtual function @var{fndecl} when constructing thunks, 9915@code{false} otherwise. By default, the hook returns @code{true} for all 9916functions, if a target supports aliases (i.e.@: defines 9917@code{ASM_OUTPUT_DEF}), @code{false} otherwise, 9918@end deftypefn 9919 9920@defmac TARGET_FORMAT_TYPES 9921If defined, this macro is the name of a global variable containing 9922target-specific format checking information for the @option{-Wformat} 9923option. The default is to have no target-specific format checks. 9924@end defmac 9925 9926@defmac TARGET_N_FORMAT_TYPES 9927If defined, this macro is the number of entries in 9928@code{TARGET_FORMAT_TYPES}. 9929@end defmac 9930 9931@deftypefn {Target Hook} bool TARGET_RELAXED_ORDERING 9932If set to @code{true}, means that the target's memory model does not 9933guarantee that loads which do not depend on one another will access 9934main memory in the order of the instruction stream; if ordering is 9935important, an explicit memory barrier must be used. This is true of 9936many recent processors which implement a policy of ``relaxed,'' 9937``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC, 9938and ia64. The default is @code{false}. 9939@end deftypefn 9940 9941@deftypefn {Target Hook} const char *TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (tree @var{typelist}, tree @var{funcdecl}, tree @var{val}) 9942If defined, this macro returns the diagnostic message when it is 9943illegal to pass argument @var{val} to function @var{funcdecl} 9944with prototype @var{typelist}. 9945@end deftypefn 9946 9947@deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (tree @var{fromtype}, tree @var{totype}) 9948If defined, this macro returns the diagnostic message when it is 9949invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL} 9950if validity should be determined by the front end. 9951@end deftypefn 9952 9953@deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, tree @var{type}) 9954If defined, this macro returns the diagnostic message when it is 9955invalid to apply operation @var{op} (where unary plus is denoted by 9956@code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL} 9957if validity should be determined by the front end. 9958@end deftypefn 9959 9960@deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, tree @var{type1}, tree @var{type2}) 9961If defined, this macro returns the diagnostic message when it is 9962invalid to apply operation @var{op} to operands of types @var{type1} 9963and @var{type2}, or @code{NULL} if validity should be determined by 9964the front end. 9965@end deftypefn 9966 9967@defmac TARGET_USE_JCR_SECTION 9968This macro determines whether to use the JCR section to register Java 9969classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both 9970SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0. 9971@end defmac 9972 9973@defmac OBJC_JBLEN 9974This macro determines the size of the objective C jump buffer for the 9975NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value. 9976@end defmac 9977